* ... . .. .
f iill~llllllliH~I~illill 'iELECTFAUG 4 199
PHASED ARRAY ANTENNA ANALYSISUSING HYBRID FINITE ELEMENT METHODS
DISSERTATION
Daniel T. McGrath, Major. USAF
AFIT/DS/ENG/93-4
(D 3Th~7fOK STTEMENT AAp?0yed ux f~z t.Jca
DEPARTMENT OF THE AIR FORCE
_ .AIR UNIVERSITY
-- IR FORCE INSTITUTE OF TECHNOLOGY
Wright-Patterson Air Force Base, Ohio
, - ii i ii8 2 i 2i
For
N~SCRA&IL)IIC TAB
AFIT/DS/ENG/93-4 .
Avd, arDliaiy Cfces
Dist Avd•j or
ýOQiCldl
DTIC QtATITAUY
PHASED ARRAY ANTENNA ANALYSISUSING HYBRID tAINITE ELEMENT METHODS
DISSERTATION
Daniel T. McGrath, Major, USAF
AFIT/DS/ENG/93-4
Approved for public release; distribution unlimited
AFIT/DS/ENG/93-4
PHASED ARRAY ANTENNA ANALYSIS
USING HYBRID FINITE ELEMENT METHODS
DISSERTATION
Presented to the Faculty of the School of Enginc.ring
of the Air Force Institute of Technology
Air University
In Partial Fulfillment of the
Requirements for the Degree of
Doctor of Philosophy
Daniel T. McGrath, B.S., M.S.
Major, USAF
June, 1993
Approved for public release; distribution unlimited
Preface
This dissertation research was in the area of computational electromagnetics, which has
the goal of producing computer codes useful for designing devices with little or no reliance on
hardware prototyping. The particular application is to design of the radiating elements of
electronically scanned phased arrays, where it is hoped that these tools will lead to improved
designs for airborne and spaceborne antennas. The work was supported by the USAF Rome
Laboratory's Electromagnetics and Reliability Directorate, Hanscom AFB, MA.
I would like to thank my advisor, Dr. Vittal Pyati, and my sponsor, Dr. Robert Mailloux,
for their confidence and encouragement. I am indebted to Doug Burkholder, Tony Schooler and
Russ Milliron of AFIT/SC for their assistance with AFIT's computers; to Michelle Champion of
Rome Laboratory for her help in acquiring and fabricating experimental hardware; to Jack
Tiffany of AFIT's machine shop, also for hardware fabrication; and to Dan Mullinix and Bob
Lindsay for help with microwave measurements. Finally, I owe the most to my wife, Victoria,
for all she has done to free my time to pursue this research, and for her patience and
understanding.
iii
Table of Contents
Preface . .................................................... iii
List of Figures . ............................................... vii
List of Tables . ................................................ x
List of Symbols ................................................ xi
Abstract . ................................................... xiii
I. Introduction and Background .. .................................... 11.1. Introduction .. ........................................ 11.2. Phased Array Antenna Electromagnetic Analysis ................. 21.3. The Need for Improved Analysis Methods ....................... 31.4. Methods in Computational Electromagnetics ..................... 6
II. Solution Overview . ........................................... 92.1. Problem Description . .................................... 92.2. The Matrix Equation .................................... 112.3. Finite Elements . ...................................... 112.4. The Weak Form Functional .............................. 122.5. Development Approach . ................................. 14
III. Interior Region Problem - Finite Element Formulation .................... 163.1. The Variational Statement ............................... 163.2. Scalar vs. Vector Finite Elements ............................ 173.3. Discretization via Galerkin's Method ......................... 203.4. Homogeneous Coordinates . ............................... 213.5. Volume Integral Computations ............................ 22
IV. Waveguide Continuity Conditions . ................................ 254.1. Combined-Source Integral Equation and Modal Expansion ........... 254.2. Discretization . ....................................... 284.3. S Parameters . ....................................... 29
V. Periodic Radiation Condition . ................................... 315.1. Periodic Integral Equation ............................... 315.2. Discretization . ....................................... 335.3. Floquet Mode Limits . .................................. 345.4. Active Reflection Coefficient and Element Radiation Pattern .......... 36
VI. Periodic Boundary Conditions . .................................. 386.1. Unit Cell Representation ................................ 386.2. Mapping from an Infinite System ............................ 406.3. Boundary Functional . ................................... 42
iv
6.4. Radiation Boundary ................................... . 446.5. Summ ary . .......................................... 45
VII. Validation - RF Device Problem . ................................ 467.1. Computer Code Implementation ............................. 46
7.1.1. General Procedure . .............................. 467.1.2. M atrix Solution . ................................ 47
7.2. Waveguide Discontinuities . ............................... 497.2.1. Iris in Coaxial Waveguide . ......................... 497.2.2. Circular Waveguide Mode Converter ................... 51
7.3. Printed Circuit Devices . ................................. 527.3.1. Microstrip Transmission Line ....................... 527.3.2. Microstrip Meander Line .......................... 55
7.4. Importance of Higher Order Modes ......................... 557.5. Summ ary . .......................................... 58
VIII. Validation - Cavity Array Problem .............................. 608.1. Computer Code Implementation ............................. 608.2. W aveguide Arrays . .................................... 60
8.2.1. Rectangular Array . .............................. 608.2.2. Rectangular Array with Conducting Iris ................. 638.2.3. Circular Waveguide Arrays . ........................ 63
8.3. Pyramidal Horn Array . ................................. 678.4. Coaxial-to-Rectangular Waveguide Launcher .................... 69
IX. Validation - General Array Problem .............................. 739.1. Computer Code Implementation ............................. 739.2. W aveguide Arrays . .................................... 749.3. Microstrip Patch Array . ................................. 769.4. Clad Monopole Array Experiment . .......................... 78
9.4.1. Initial Validation ............................... 799.4.2. Bandwidth Enhancement due to Cladding ................ 799.4.3. Experiment . .................................. 82
9.5. Printed Dipole Radiator . ................................. 839.5.1. Element Design . ............................... 839.5.2. Calculations . .................................. 86
9.6. Flared Notch Radiator . .................................. 879.6.1. Element Design ............................... 879.6.2. Array Performance Calculations ..................... 88
9.7. Sum m ary . .......................................... 90
X. Conclusions and Recommendations ................................. 9310.1. Conclusions . ....................................... 93
10.1.1. Theory and Formulation ......................... 9310.1.2. Implementation . ............................... 9410.1.3. Validation . .................................. 94
10.2. Recommendations . .................................... 95
v
Appendix A: The Electric Field Functional .............................. 97A. 1. Variational Principle vs. Weak Form ........................ 97A.2. The Adjoint Problem . .................................. 98A.3. Continuity Conditions for Waveguide Apertures ................. 101A.4. Galerkin's Method vs. Rayleigh-Ritz ........................ 102
Appendix B: Waveguide Mode Function Inner Products ..................... 104B. 1. Approach ......................................... 104B.2. Rectangular W aveguide ................................. 105B.3. Gaussian Quadrature Integration ........................... 108
Appendix C: The Periodic Integral Equation ............................ 110C. 1. The MFIE for Planar Current Sources ....................... 110C.2. The Periodic Magnetic Field Integral Equation .................. 112C.3. Skewed Array Lattices . ................................. 116C.4. Expansion Function Fourier Transform ....................... 118C.5. Integral Equation from Floquet Modes ....................... 120
Appendix D: Periodic Boundary Conditions for the Finite Element Problem ........ 123
References . ................................................. 128
V ita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ....... 133
vi
List of Figures
1. Flared Notch Radiator: (a) Printed Circuit Fed from Coaxial Waveguide;(b) Geometry Model for Method of Moments ........................ 4
2. Printed Dipole Radiator: (a) Actual Geometry with Microstrip Balun andCoaxial Feed; (b) Method of Moments Model ....................... 5
3. Stacked Patch Radiator with Coaxial Feed: (a) Two Continuous Substrate
Layers; (b) Non-continuous Top Substrate .......................... 5
4. General Phased Array Radiator Problem .............................. 9
5. Unit Cells (Stacked Patch Arrays): (a) Rectangular Lattice;(b) Triangular Lattice . ...................................... 10
6. Subdivision of a Volume Region into Tetrahedra (one quadrant of a disk):(a) Interior Edges Visible; (b) Interior Edges Hidden ................... 12
7. A Generic Cavity Array Problem .................................. 14
8. A Generic Passive RF Device Problem ............................. 15
9. A Three-Dimensional Linear Expansion Function: (left) Eight TetrahedraSurrounding a Node; (right) Linear Finite Element with Surfaces ofConstant Function Value . .................................... 17
10. A Two-Dimensional Vector Finite Element: (a) Direction (arrow lengthindicates function magnitude); (b) Constituent Linear Scalar FiniteElem ents . ............................................. 19
11. Equivalence Model for Waveguide/Cavity Problem: (a) Original Problem;(b) Interior Equivalent; (c) Exterior Equivalent ...................... 26
12. Infinite Array of Apertures, Skewed Lattice .......................... 32
13. Typical Scalar Finite Element Fourier Transform (Contours in dB, samplepoints are Floquet modes for a square lattice with hL=hy=dx/5=dy/5) ....... 35
14. Unit Cell Definitions: (a) Circular Waveguide Array; (b) RectangularPatch Array . ............................................ 39
15. Example Unit Cell Mesh Showing Opposite Faces ...................... 40
16. Triangular Finite Element Mesh for Infinite Two-Dimensional PeriodicProblem .. .............................................. 41
vii
17. Periodic Boundary Condition Algorithm .............................. 42
18. Solution Procedure in Program TWOPORT .......................... 47
19. Comparison of Solution Times for LU Decomposition (IMSL) and ConjugateGradient Method (CPU seconds on VAXO 8650 minicomputer) ............. 49
20. Convergence of Residual Norm and Reflection Coefficient using ConjugateGradient M atrix Solver ..................................... 50
21. Finite Element Mesh for Coaxial Waveguide Section with Conducting Iris ....... 50
22. Measured and Computed S I Magnitude and Phase for Coaxial Iris ........... 51
23. Finite Element Meshes for Circular Waveguide Step Discontinuities (ModeConverters) with a2 =1.4": (a) a,= 1.05"; (b) a,= 1.15".................. 53
24. Mode Conversion Ratio for Circular Waveguide Mode Converters: TWOPORTand Multimode [35] Calculations ............................... 53
25. Microstrip Transmission Line Section in Conducting Enclosure ............... 54
26. Microstrip Meander Line: (a) Wafer Metallization Dimensions (in Jim);and (b) Finite Element Mesh for Substrate ........................... 56
27. Comparison of Measured and Calculated S2, Phase for Microstrip Meander Line... 56
28. Reflection Coefficient for One-Quadrant Coaxial Waveguide Iris Calculatedwith Varying Numbers of Higher-Order Modes ....................... 57
29. Solution Procedure in Progcaia CAVIARR ............................ 61
30. Rectangular Waveguide Test Case: (a) Array Lattice; and (b) Tetrahedron Mesh . . 62
31. H-Plane (0=0) Active Element Gain vs. Scan Angle for RectangularW aveguide Test Case ...................................... 62
32. Active Reflection Coefficient vs. Scan Angle for Rectangular WaveguidePhased Array with Conducting Irises .............................. 64
33. Circular Waveguide Array Test Case: (a) Lattice; and (b) Tetrahedron Mesh ..... 64
34. Circular Waveguide Array Active Reflection Coefficient vs. Scan Angle ......... 65
35. Active Reflection Coefficient vs. Scan Angle - Circular Waveguide Arraywith and without Dielectric Loading ............................. 66
viii
36. Active Reflection Coefficient vs. Scan Angle - Circular Waveguide Array
with and without Dielectric Plugs .............................. . 67
37. Pyramidal Horn Radiator: (a) Dimensions; (b) Tetrahedron Mesh ............ .68
38. Active Element Gain vs. E-Plane Scan Angle for Pyramidal Horn Array(Measured Data from Amitay & Gans 1421) ......................... 69
39. End-Wall Transition from Coax to Rectangular Waveguide (After Tang &Wong [431): (a) Geometry; (b) Cutaway of Tetrahedron Mesh ............ 70
40. Reflection Coefficient (S1 j) for Coax-Rectangular Launcher:Probe Length=9mm; Post Height=5mm; X-Band Waveguide ............ 70
41. Active Reflection Coefficient for Rectangular Waveguide Arraywith Coaxial Launcher and Rectangular Waveguide Feeds ............... 71
42. Solution Procedure in Program PARANA ............................ 74
43. Finite Element Mesh for a Skewed-Lattice, Circular Waveguide Array Unit Cell . . . 75
44. Circular Waveguide Array Active Reflection Coefficient - Comparison ofResults Uscig Cavity Array (CAVIARR) and General Array (PARANA) Models . 76
45. Rectangular Waveguide Array Active Reflection - Comparison of ResultsUsing Cavity Array (CAVIARR) and General Array (PARANA) Models ..... 77
46. Microstrip Patch Radiator (dimensions in wavelengths) ................... 77
47. Active Reflection Coefficient for Microstrip Patch Array, E-Plane Scan ........ 78
48. Reflection Coefficient vs. Scan Angle (q5=0 plane) for a Monopole Array ....... 80
49. Finite Element Mesh for a Clad Monopole in a Triangular Lattice ............ 80
50. Active Reflection Coefficient Magnitude vs. Scan Angle (horizontal axis, deg.)and Frequency (vertical axis, GHz) for Bare (top) arld Clad (bottom) Monopoles . 81
51. Experimental Array Geometry and Elements used for Coupling Measurements .... 82
52. Measured and Computed Active Reflection Coefficient vs. Angle forClad Monopole Array Experiment ............................... 84
53. Printed Dipole Radiator Design . ................................. 85
54. Exploded Finite Element Mesh for Printed Dipole Radiator ................. 85
ix
55. Computed Active Reflection Cc,..,hcient at Broadside Scan for Reduced-Height(12.5mam) Printed Dipe!, "or Two Dipole Lengths ...................... 86
56. Computed Active Reflection Coefficient (Normalized) vs. Scan Angle for
Reduced-Height (12.5mm) Printed Dipole, 4.8 GHz ..................... 87
57. Flarod Notch Element Design .................................. 88
58. Finite Element Mesh for Flared Notch Radiator (Rectangular Lattice):(a) Unit Cell Showing Coax Aperture; (b) Substrate Surface Mesh ........... 89
59. Active Reflection Coefficient vs. Frequency for Printed Flared Notch Arrays .... 90
60. Active Reflection Coefficient vs. Scan Angle, Printed Flared Notch inRectangular Lattice: (a) E-Plane Scan; (b) H-Plane Scan ................. 91
BI. Procedure for Matrix Fill Calculations Involving Waveguide Modes .......... 105
Cl. Two Dimensional Convolution of Unit Cell Field with DiracIm pulse Sequence ........................................ 113
D 1. Periodic Functions ......................................... 123
D2. Expansion/Weighting Functions ..................................... 124
D3. "Wrapped" Domain ............................................. 126
List of Tables
I. Mesh Sizes, Iterations and Matrix Solve Time for TWOPORT Test Cases(Solve time is CPU time on VAX® 8650 minicomputer) ................. 58
II. Mesh Size, Execution Time and Matrix Storage for PARANA Test Cases ....... 92
x
List of Symbols
Symbol Most Frequent Use. type 0 and MKS dimensions H
a,b Rectangular waveguide dimensions in x and y, respectively
or coaxial waveguide inner and outer radii, respectivy (real constants) [m]
dx,dy Array unit cell dimensions in x and y, respectively (real constants) [m]
F. Electric field (vector, complex, time harmonic function) [Volts/m 2]
E• Approximation of E by finite elements
Orthonormai waveguide eigenfunction for mode i (vector, real)
H Magnetic field (vector, complex, time harmonic function) [Amps/m 2]
T Electric current (vector, complex, time harmonic function) [Amps/m]
k Wavenumber, k=27r/X (real constant) [I/m]
kx,ky Spatial frequency coordinates parallel to x and y, respectively
Mý1 Magnetic current (vector, complex, time harmonic function) [Volts/m]
N Number of finite element mesh edges in interior region (integer), also the numberof unknowns in the matrix equation
b(x,y) Dirac delta, =0 when x;40 or y;60
bij Kronecker delta, =0 when i;•j
"y Array lattice skew angle (real constant)
rA,rB Waveguide apertures (boundary surface)
rR Radiation boundary (surface)
Pr_,r+,,r y,r+y Unit cell side wall boundaries (surfaces)
E Permittivity (material physical property, complex scalar) [Farads/m]
Co Permittivity of free space (real, constant) [Farads/m]
En• Neumann's number, = 1 for n = 0; = 2 for n l I
SImpedance (material physical property, comp!ex scalar) [Ohms]
xi
To Impedance of free space (real, constant) [Ohms]
0 Spherical coordinate angle measured from the +z axis [radians]
00 Array scan angle (real constant) [radians]
X Wavelength (real constant) [m)
SPermeability (material physical property, complex scalar) [Henrys/m]
AO Permeability of free space (real, constant) [Henrys/m]
Two-dimensional Fourier transform of component of expansion function for edge stangential to radiation boundary
Spherical coordinate angle measured from +x axis [radians]
0o Array scan angle (real constant) [radians]
,y Beam steering phase shifts (real constants) [radians]
•. Vector expansion function associated with mesh edge s
',Si Inner product of expansion function ý. and waveguide mode function gi
al Volume region, usually the unit cell excluding interiors of conductors
W Radian frequency (real constant) [radians/s]
xii
Abstract
This research in computational electromagnetics addressed the problem of predicting the
near-field mutual coupling effects in phased array antennas. It developed and demonstrated a new
analysis technique that uses the finite element method (FEM) in combination with integral
equations. Due to FEM's inherent ability to model inhomogeneous dielectrics, the new capability
encompasses many radiator types that were not amenable to analysis by previously-existing
methods. The analysis considers the general case of a radiator in an infinite array that is fed
through a ground plane by one of three types of waveguides: rectangular; circular; or circular
coaxial. Accurate feed modeling is accomplished by enforcing continuity, between the FEM
solution and an arbitrary number of waveguide modes, across the ground plane aperture. A
periodic integral equation is imposed at a plane above the antenna's physical structure to enforce
the radiation condition and to confine the analysis to a single array unit cell. The electric field
is expanded in terms of vector finite elements, and Galerkin's method is used to write the
problem as a matrix equation. The Floquet condition is imposed as a transformation of the
matrix, which is equivalent to wrapping opposing unit cell side walls onto each other with a plase
shift appropriate to the scan angle and lattice spacings. The solution of the linear system,
accomplished using the conjugate gradient method, gives the electric field, from which the active
reflection coefficient and active element gain are calculated.
The theory and formulation were used to develop a general-purpose computer code. The
use of commercial CAD (computer-aided design) software for geometry and mesh generation
makes the code geometry independent. It was validated by comparing its results to published data
for arrays of open-ended waveguides, monopoles and microstrip patches. Predictions for
dielectric-clad monopoles were validated by a hardware experiment. Finally, the code was used
to predict the scanning properties of arrays of printed dipoles and printed flared notches.
xiii
PHASED ARRAY ANTENNA ANALYSIS
USING HYBRID FINITE ELEMENT METHODS
L Introduction and Background
1.1. Introduction
A critical problem in phased array antenna design is that of controlling the mutual cou-
pling between individual antennas, or radiators, that comprise the array. Mutual coupling may
reduce antenna efficiency by creating a reflection mechanism that depends on both the radiator
geometry and the scan angle. The limiting case, but one that frequently results from poor
radiator design is scan blindness, meaning that there are one or more angles in the desired field
of view where the reflection is total. Since mutual coupling is inherently a near-field electromag-
netic effect, it is rarely possible to achieve an acceptable radiator design without the ability to
accurately model the near fields. As we attempt to design phased array antennas for new applica-
tions, we often find that existing field computation techniques, most notably moment methods,
do not adequately account for the radiator's topology, feed structure, or dielectric materials. This
is especially true for several broadband radiators that incorporate dielectrics either as structural
support or as electrical loading that reduces their size. Therefore, improved computation tech-
niques are required for use as design tools for broadband phased array antennas.
The objective of this dissertation research was to develop and demonstrate a new analysis
method versatile enough to predict the performance of a variety of radiators. The method is a
hybrid of the finite element method (FEM) with integral equation continuity conditions. The
integral equation for one of three types of waveguides (a summation over dominant and higher-
, . .i | I II
order waveguide modes) provides an accurate method for including feed structure effects. The
use of a periodic integral equation to represent the fields above the array enforces the radiation
condition and makes the analysis tractable by confining it to a single array unit cell. A new
periodic boundary condition for finite elements was derived to account for the mutual coupling
across unit cell side walls. The work culminated in a general-purpose computer code that suc-
cessfully predicted the scan-dependent properties of a variety of arrays such as open-ended
waveguides, microstrip patches and printed flared notches. When possible, the results were
confirmed by comparisons to data in the scientific literature that was obtained from either mea-
surements or by other methods of calculation. These validation cases are only a sampling of the
capability of this new analysis tool, which is able to predict the scanning performance of most
array radiators that are in use or proposed for use. Its future use as a design tool is bound to
improve the performance of phased array antennas.
1.2. Phased Array Antenna Electromagnetic Analysis
The most important reason for near-field electromagnetic analysis of phased arrays is
impedance matching, which translates directly into radiation efficiency and low VSWR (voltage
standing wave ratio) to prevent damage to transmitter components in high-power applications.
Paradoxically, radiating elements that are well matched in isolation do not necessarily remain so
when they are arranged in a closely-spaced lattice to form an array. Taking one radiator to be
a reference element, some of its transmit power couples directly into other elements, and some
power from each of the other elements will likewise couple directly into the reference element.
The power returning to the transmitter by way of mutual coupling represents a reflection mecha-
nism. In order to achieve a good active impedance match, each element must, by itself, be
slightly mismatched, so that its self-reflection vectorially cancels the sum of couplings from other
elements. There are numerous methods for controlling these mismatches, such as altering the
2
radiator geometry or including matching circuits in the feed network, but they can only be
effective if an accurate solution for the mutual coupling is available [1:16621.
Most existing solutions for array mutual coupling use the method of moments (MoM) in
conjunction with an infinite array approximation. The approximation makes the problem compu-
tationally tractable since periodicity conditions may be used to restrict the analysis to the space
around a single radiator, called a unit cell. It is a reasonable approximation for large arrays, and
it is also used as part of the present method.
1.3. The Need for Improved Analysis Methods
There are several trends in antenna development that are causing array element designs
tc outstrip the available analysis methods. One is the desire to use electronic scanning antennas
in applications such as airborne satellite communications and airborne surveillance radar, in which
the antennas' intrusion, protrusion and weight must be minimized [21. Another is the growing
trend toward millimeter wave frequencies, leading to dimensions so small that it is impractical
to fabricate and assemble and array one element at a time. Yet another is the potential cost
reduction of using printed circuit and integrated circuit, or monolithic, fabrication [3].
The radiating elements that are proposed to meet these new requirements often include
irregularly-shaped conducting surfaces in combination with inhomogeneous dielectrics. An
example, shown in Figure la, is the "flared-notch" element [4]. The dielectric card is clad with
metal on the back side except for a slot that opens progressively wider near the top. The front
side is bare except for a microstrip feed line. It is shown here fed from a coaxial transmission
line that penetrates a ground plane, but the feed line could be microstrip as well. A recent MoM
analysis of this radiating element in the phased array environment used the simplification shown
in Figure lb: the dielectric is ignored; and the feed line is replaced by an delta-gap source across
the slot 151. It is clear that this model cannot predict the effects of high-dielectric-constant sub-
3
DIELECTRICFLARED NOTCH
DIDELTA GAPSOURCE
GROUN6PLANE
COAX-MICROST RIP-SLOTLINE FEED
(a) (b)
Figure 1. Flared Notch Radiator: (a) Printed Circuit Fed from Coaxial Waveguide;(b) Geometry Model for Method of Moments
strates (monolithic antennas commonly use Gallium Arsenide, whose relative permittivity is 12.8-
12.9) or of radiation from the feed structure.
A second example, shown in Figure 2a is a dipole radiator metallized on one side of a
dielectric card, with a balun feed metallized on the other [61. The dielectric may be trimmed or
notched at each end of the dipole to reduce the mutual coupling between elements located at
intervals along the card. Figure 2b is the structure actually modeled using an innovative MoM
approach. In this case, the presence of the dielectric was taken into account by using a Green's
function for a parallel-plate region periodically loaded with dielectric slabs 171,18]. Some results
of that work confirm that the dielectric has a pronounced effect on the wide angle scanning
properties. On the other hand, the sweep-back of the dipole arms, which is known to be impor-
tant for achieving a good impedance match at wide scan angles 191, is neglected. Also, the feed
is modeled as a simple delta-gap source, neglecting the effects of the balun.
4
DIELECTRIC R
RADIATOR DELTA GAPA 'SOURCE
FEED LINEAND BALUN -
GROUND Ž" -
PLANE - >
COAXIAL FEED
(a) (b)
Figure 2. Printed Dipole Radiator: (a) Actual Geometry with MicrostripBalun and Coaxial Feed; (b) Method of Moments Model
PARASITIC ELEMENT SUBSTRATE LAYERS• llll• aPATCH RADIATOR __•
v (a)
COAXIAL FEED GROUND PLANE
(b)
Figure 3. Stacked Patch Radiator with Coaxial Feed: (a) Two ContinuousSubstrate Layers; (b) Non-continuous Top Substrate
5
A third and final example is shown in Figure 3a, It is essentially a rectangular microstrip
patch, fed through the ground plane from a coaxial cable. The second "proximity-coupled" patch
on top of the upper dielectric layer is intended to increase the overall bandwidth beyond the 5%
[101 that is typical of a single patch. The array properties of this radiating element have been
predicted using MoM Il1], with the coaxial feed represented by a frill current source. The mo-
ment method approach for this, and other microstrip problems, relies on using Green's functions
for layered, infinite dielectric slabs. Those Green's functions do not apply to situations such as
Figure 3b, in which one or both substrates are not continuous layers.
These three examples illustrate the deficiencies in previous mutual coupling computation
techniques; and the corresponding new capabilities that have been obtained with the hybrid finite
element method: (1) multiple dielectrics with arbitrary shape; (2) irregular conducting surfaces
that support currents flowing in arbitrary directions; and (3) detailed feed structures. A further
objective that is also important is geometry independence: Whereas each of the three examples
discussed above used a specialization of MoM to the particular structure and required develop-
ment of a separate computer code, the present work resulted in a code that can model all three,
and many others as well.
1.4. Methods in Computational Electromagnetics
The techniques of "classical electromagnetics" provide formalisms for casting physical
problems as mathematical boundary value problems. Since the solutions that may be obtained
by the purely analytical methods are restricted to canonical geometries, most electromagnetic
design problems of current interest require the use of numerical methods to obtain a solution.
The techniques of "computational electromagnetics" (CEM) are formalisms for mapping boundary
value problems from continuous to discrete forms so that they may be solved by computer.
Therefore, the objective of CEM is to produce tools, i.e. computer codes with which device
6
designs may be evaluated without resorting to hardware experiments. The tools must have four
essential characteristics: effectiveness, reliability, efficiency, and versatility. In other words, they
must consistently obtain correct results at reasonable cost for a variety of problem geometries.
The previous section showed that the shortcomings of current methods for phased array near-field
analysis are mainly in effectiveness (due to simplifications of the actual problem geometry) and
versatility (due to the restriction of each to very specialized geometries).
The requirement for effectiveness limits the search for new techniques to integral equation
(MoM) and partial differential equation (PDE, finite element and finite difference) methods. The
former is by far the most well advanced for solving antenna problems because the integral
equations incorporate Green's functions that satisfy radiation conditions, forcing all valid solutions
to decay to zero with increasing distance from the field sources (equivalent currents). The PDE
techniques, on the other hand, more easily account for dielectric inhomogeneities for which
Green's functions are not available. The finite element method is more appropriate for devices
with irregular, especially curved surfaces, because it may use irregular grids, or "meshes," while
the finite difference method typically uses regular, Cartesian grids. A similar need to model
objects that include inhomogeneous dielectrics has led researchers in electromagnetic scattering
to consider hybrids of the finite element method with integral equation methods [121-f14]. By
surrounding the object with an imaginary boundary in free space surrounding the scatterer, the
finite element method may be used to solve for the fields inside the boundary as though it were
an enclosed region, and an integral equation is imposed to ensure field continuity across the
boundary. Hence, the hybrid finite element method (HFEM) appeared to be a likely choice for
the phased array radiator problem as well, provided that a means could be found for implement-
ing periodic boundary conditions. The success of that implementation and a demonstration of its
benefits are some of the important results of this dissertation research.
7
The next chapter will pi -,ent an overview of the solution approach, which involves three
novel continuity conditions for the three dimensional finite element problem. The detailed analy-
ses and derivations constituting the problem "formulation" (its description as a mathematical
boundary value problem, and its reduction to a linear system of equations ) are given in Chapters
III-VI. They are intended as documentation, or as a trail for the reader who would attempt a
similar solution to related electromagnetic problems. They are not essential to understanding the
results of validation tests and hardware experiments presented in Chapters VII-X.
This hybrid finite element method is a frequency-domain approach. Hence, throughout
this document, all field and current quantities are understood to be time-harmonic, with the
complex exponential eJt time dependence suppressed.
8
II. Solution Overview
2. 1. Problem Description
The generic problem geometry is shown (cross-section) in Figure 4. The "interior
region," denoted 0), is a section of an array unit cell. It is bounded by the surface denoted r',
whose bottom wall is the ground plane plus the feed waveguide aperture. The top wall, called
the "radiation boundary," is an imaginary constant-z surface at an arbitrary position in free space
above the radiator structure. The side walls conform to the unit cell boundaries. For example,
Figure 5 shows two arrays (stacked-patch radiators), one with a rectangular lattice and the other
with a triangular or "skewed" lattice. In each case the unit cell is a cylinder extending
indefinitely in the +z directions. Its side boundaries are chosen to satisfy the periodicity
conditions (discussed in Chapter VI and Appendix C). The region Q is formed by simply
truncating the unit cell at some plane above the array.
z
INFINITE ARRAY
I RADIATION BOUNDARY---------- ---- ----------. . . . .. . . . . . . . .- -_- -. . . . .i . . . . z " h
SIDE WALL
UNIT CELL agoGROUND z 0
PLANE
DIELECTRICS CONDUCTORS
WAVEGUIDE APERTURE
Figure 4. General Phased Array Radiator Problem
9
UNIT CELL
DIELECTRICLAYERS
x GROUNDyPLANE
UNIT CELL
DIELECTRI
LAYERS
xKGROUND
LAYERSLANE
Figure 5. Unit Cells (Stacked Patch Arrays): (a) Rectangular Lattice,(b) Skewed (Triangular) Lattice
10
1) may now be viewed as a cavity containing any number of material regions, each with
distinct constitutive parameters E and 1L, which may be complex (lossy). This research will only
consider linear and isotropic materials. There may also be voids within 0 that represent the
interiors of T rfectly conducting obstacles. Infinitely thin wires and open surfaces such as patches
and strips are also permitted.
2.2. The Matrix Equation
The solution to the boundary value problem represented by Figure 4 is the electric field
E(x,y,z) everywhere inside 9 and on F. HFEM will find an approximation to E in terms of
piecewise-continuous expansion functions weighted by a column vector of coefficients, denoted
E. Those coefficients are the c,!uti'n to the matrix equation
[ SEE , SEJ ]E = E (I)
Complete explanations of the terms in this equation are given in succeeding chapters, but briefly:
The matrix SEE is sparse, representing local interactions between field sources inside g'; SEJ
represents interactions between field sources on the nonconducting parts of r through integral
equations. The right side vector Einc is due to a field incident on F from the feed waveguide.
The performance parameters that are of greatest interest are the active reflection coefficient and
active element gain, which may be found directly from those parts of E on the waveguide
aperture and radiation boundary, respectively.
2.3. Finite Elements
The region 01 will be subdivided into small volume elements. Four-sided tetrahedra were
chosen because they conform more readily to irregular and curved surfaces than other popular
choices such as six-sided "bricks." The volume elements are often referred to as cells, and their
four vertices as nodes. The collection of tetrahedra is called the mesh. Material properties will
[ "l
be assumed constant within each cell. Figure 6 is an example mesh (one quadrant of a thick disk)
that illustrates an important flexibility of tetrahedron meshes: the mesh density may be varied
within an object. Although 10-20 nodes per linear wavelength is usually an adequate sampling
rate, one may wish to sample finer in regions where the field is expected to have singularities;
or in order to capture fine details of an object's geometry.
Finite elements are polynomial functions that are defined over individual cells, or sub-
domains. Chapter III will give a more detailed description of the linear, vector finite elements
used in this work. These functions are used as expansion and testing functions for Galerkin's
method, which is the mechanism used to reduce the boundary value problem to a matrix problem.
2.4. The Weak Form Functional
The principal distinction between finite element and moment methods (as the terms are
commonly used within the electromagnetic research community) is that the former is applied to
variational statements, while the latter is applied to integral equations f15:161. The variational
(a) (b)
Figure 6. Subdivision of a Volume Region into Tetrahedra (one quadrant of a disk):(a) Interior Edges Visible; (b) Interior Edges Hidden
12
statement used here is the weak form of the vector wave equation. The time-harmonic form of
the wave equation for electric fields in a source-free, inhomogeneous region is:
VXIVXE-kArE = 0 (2)Ar
A functional is constructed by taking its inner product with a trial function, W, then applying a
Green's identity (see Appendix A):
F(E= J Vx *.vxW E dv + -kOlioJW*.Jds=0 (3)
This is called a weak form because the Green's identity has shifted one derivative from the field
E to the trial function W, thus weakening the differentiability requirement on E. This functional
has three difficulties that integral equations do not:
(a) the Helmholtz equation specifies the curl only and not the divergence. There-
fore, extra effort is required to enforce the divergence condition V- (EE)=0.
(b) Boundary conditions are not included. Thus, although (3) is not restricted to
a class of problems with uniform boundary conditions, extra effort is required to
ensure the satisfaction of all boundary conditions that are present.
(c) The radiation condition is not enforced.
Chapter III will discuss how (a) and (b) are resolved by choosing vector expansion functions that
obey the divergence condition and that satisfy boundary conditions at both j,• w, 1 conductors and
dielectric interfaces. Chapters IV and V show how the radiation condmin , 1 rced by substi-
tuting an integral equation for J into the boundary integral of (3). In t, ,t the waveguide
aperture, that integral equation will take the form of a sum over waveguide modes. In the case
of the radiation boundary, it will be a periodic integral equation. which may be written as a sum
over spectral domain sample points, i.e. Floquet modes.
13
The last detail to be addressed is the problem of enforcing periodicity conditions at the
unit cell walls. The implementation is straightforward in theoretical terms. It has been accom-
plished by others for two-dimensional grating problems [ 161. Chapter VI discusses its extension
to three dimensions and the means for implementing it algorithmically: The matrix is first
constructed as though the unit cell walls are open circuit boundaries; then the matrix is modified,
creating some new terms and removing others. The effect is as if opposing mesh side walls were
folded around onto each other with a phase shift appropriate for the array scan angle and the unit
cell dimensions.
2.5. Development Approach
Figure 7 illustrates a generic "cavity array problem," a somewhat simpler problem than
the "general array" problem of Figure 4. It is still a phased array antenna, but the radiators are
separated from each other by conducting walls. Their mutual coupling is only through apertures
in a conducting ground plane. This is appropriate to a restricted class of radiators such as open-
zUNIT CELL
VR' BOUNDARY
z-d ----- GROUND PLANE
PERFECTLY\//_=- CONDUCTING
/ CAVITY
WEGUIDE ,,DIELECTRICS, AND/OR
r•W CONDUCTORS
Figure 7. A Generic Cavity Array Problem
14
ended waveguides, horns and slots. This problem embodies all the same aspects as the general
array problem except for the periodicity conditions on the unit cell side walls.
Figure 8 is a further simplification, which will be referred to as the "RF device problem."
The interior region is again a cavity with perfectly conducting walls, as in Figure 7. But here,
both inlet and outlet apertures lead into waveguides. This embodies all of the aspects of the
cavity array problem except the periodic integral equation.
The RF device problem was the first stage of this research. It provided validation of the
approach and implementation for the 3D finite elements and the waveguide aperture continuity
conditions. A detailed summary of that work is given in a separate report [17]. The second
research stage replaced the outlet waveguide modes with Floquet modes in order to solve the
cavity array problem. The third and final stage in the algorithm and code development included
the periodicity conditions needed for the general array problem. Each of tie three solutions was
validated by comparing computations to results published by other authors, obtained by methods
other than FEM (measurements, mode matching, method of moments, etc.).
FF"rA _ _ _
WAVEGUIDE A WAVEGUIDE BCAVITY
z=O z-d
Figure 8. A Generic Passive RF Device Problem
15
Il. Interior Region Problem - Finite Element Formulation
The first part of the problem formulation is the application of the finite element method
to the interior region 17. This will ignore, for the time being, the field continuity conditions on
its enclosing boundary. This chapter will discuss the variational form of the problem and the
restrictions it imposes on the electric field approximation. It then discusses the nature of the
vector finite elements and shows that they satisfy the restrictions. Finally, it gives the derivation
of the interior matrix terms using Galerkin's method, and their reduction to algebraic expressions
using coordinate transformations local to each tetrahedron.
3.1. The Variational Statement
The boundary value problem consists of the operator equation (the vector wave equation),
boundary conditions and applied forces. The solution is a function, the electric field, defined
throughout Q. The finite element method attempts to solve a variational equivalent of the
problem.
The variational statement consists of a functional (usually an integral containing the
unknown function in the integrand) and admissibility restrictions on the function [ 18]. Admissible
functions are those that are in the domain of the functional and satisfy the boundary conditions.
Appendix A discusses the two forms of functionals commonly used for vector electromagnetic
problems and gives the rational for selecting the weak form (3).
There are three admissibility restrictions. First, the divergence condition V, (EE)=O
is necessary to ensure a unique solution since the operator equation only specifies the curl of E.
Second, the tangential electric field must vanish at the surface of perfect conductors, i.e.
fi×E=0 . Last, at interfaces between dielectrics, tangential E is continuous, but normal E is
discontinuous, i.e. ft'(f 1E)=d '(f 2E2) . It will be shown that these three restrictions are
16
satisfied through a careful choice of the expansion functions used to approximate E.
3.2. Scalar vs. Vector Finite Elements
The most conventional finite elements are linear functions defined relative to the mesh
nodes. Within a single tetrahedron there are four such functions, one per node. Each finite
element is defined within a single tetrahedron and is zero everywhere else. Scalar node-based
expansion functions for E are assembled from the finite elements. For example, in Figure 9,
there are eight tetrahedra surrounding the central node. The scalar expansion function is defined
over all eight cells, is equal to 1 at the center node, and goes linearly to zero at all surrounding
nodes. In order to represent a vector field, one choice is to expand it in terms of these scalar
functions with vector coefficients:
MEt E -e, s s(x,y, z) (4)
s=1
0.01,
.25
4t = .50 "
.75
1.0
"NODE S
Figure 9. A Three-Dimensional Liiear Expansion Function: (left) Eight Tetrahedra Surroun-ding a Node; (right) Linear Finite Element with Surfaces of Constant Function Value
17
where M is the total number of nodes in the mesh. However, this expansion has three important
disadvantages:
(a) The boundary conditions at perfect conductors are difficult to enforce, espe-
cially at edges and tips where the surface normal is undefined.
(b) At dielectric interfaces where E is discontinuous, the electric field normal to
the interface is discontinuous, while the tangential components are continuous.
But if a node s is on such an interface, (4) implies that all components are contin-
uous. Hence, a node-based formulation will not accurately predict the field
behavior at dielectric boundaries [19].
(c) It does not generally satisfy the divergence condition. Failure to enforce the
condition will lead to spurious non-physical solutions 1201. It has been widely
presumed that the penalty function method could be used, but Boyse et. al. point
out that penalty methods are only justified for positive definite functionals [211,
and (3) is indefinite.
Many of these difficulties can be circumvented by using vector finite elements in an
expansion of the form
N£ = e.,ý(xy,z) (5)
s=1
where now s is an edge index and N is the number of edges in the mesh. The particular form
of ý, sometimes attributed to Nedelec 1221 that has been used most successfully is 1231,1241
4s = Lifi Vfj -4 VA ) (6)
where fi and f are the linear scalar finite elements defined for the nodes i and j bounding edge
s. Lii is the length of the edge, and is included as a scaling to ensure that the component of .
tangential to the edge is a unit vector. Figure 10 illustrates the two dimensional version of this
18
2 FUNCTION DEFINEDRELATIVE TO THIS EDGEf
3 -Vf 2
(a) (b)
Figure 10. A Two-Dimensional Vector Finite Element: (a) Direction (arrow length) indicates
function magnitude); (b) Constituent Linear Scalar Finite Elements
vector function and the scalar finite elements from which it is constructed. This choice resolves
all three of the difficulties associated with the node based formulation since: (a) it allows a
simple yet effective means for imposing conductor boundary conditions; (b) it enforces continuity
of tangential field, but allows the normal field to be discontinuous at element boundaries; and (c)
it is free of divergence since
(7)-fV 2f'.-f'.V2f,=0
(fi and fj are linear, so their second derivatives are zero).
The principal disadvantages of the vector elements are algorithmic: (a) most CAD
software generates a node listing that must now be converted to an edge listing; (b) the direction
of each edge vector must be accounted for; and (c) the calculated fields must be converted back
to vector components for output or display. One further disadvantage that has been cited (191,
19
[231 is that the number of unknowns is larger since there are typically 4-5 times as many edges
as nodes in a tetrahedron mesh. Hence, there may be 1/3 - 2/3 more unknowns. This estimate is
excessive for three reasons: First, it assumes three unknowns per node, but four are actually
necessary to achieve an effective node-based formulation, using vector and scalar potentials [251.
Second, there are no unknowns associated with edges on perfect conductors, since the tangential
electric field must be zero, whereas all three components of the electric field at a node on a
conductor could be nonzero. Third, the node-based formulation may require finer sampling near
conducting edges and corners to compensate for the uncertainty in the direction of fi.
Furthermore, the connectivity between edges is lower than for nodes, typically by a factor of 2,
and hence the number of matrix entries is smaller by the same factor.
Given the above facts, the vector finite elements are clearly the better choice. The next
section will show the derivation for the interior matrix terms using the expansion (5) in
conjunction with Galerkin's method.
3.3. Discretization via Galerkin's Method
Galerkin's method is a specialization of weighted residuals, in which the trial functions
are the same as the expansion functions. Its use is permitted when the expansion functions are
in the admissible space of both the direct and the adjoint problem, as discussed in Appendix A.
Substitution of the series expansion for electric field into the operator equation leaves a
residual error R = L(E)-L(E) where E is the series approximation to E and L is the linear
operator Vx #r1 VX + r6r]" The inner product of R with a trial function W is
jRW*dv = F(E)- IV e[-J V×X V XW*(n t• (8)
EAr 'W*jdv +jkoI o jJ J -W*dsF
This includes the original functional since (L(E),W)=F(E), but since F(E)=O from (3). The
20
weighted residuals procedure forces (R,W)=0 in order to solve for the coefficients e;, giving
0= W (9)t= 1
0l P
Substituting each ý., one at a time, for W gives N equations:
et[r I-VXý-k t d k~oJ d (10)
The order of summation and integration in (10) n.ay be reversed since the coefficients e; are finite
and the functions it and Vx~t are bounded. Then (1( defines a system of N equations in the
N unknowns et. The volume integral terms are the entries in the matrix SEE from (1):
SEE ['- V Xý (11)'_ rOs,
The expansion and testing functions are each defined only on the collection of tetrahedra adjacent
to the corresponding mesh edges. Hence the integration is over f1st, the collection of cells shared
by edges s and t. These matrix equation terms will be computed by carrying out the integrations
in (11) analytically using a transformation to homogeneous coordinates.
3.4. Homogeneous Coordinates
The homogeneous or simplex coordinates are defined locally within each tetrahedron
126:266-2741. There are four coordinates t I , t2, t 3 and t4 , but one of the four can always be
eliminated using the relationship t1 +t 2 +t 3 +t 4 = 1 . The coordinate ti of a point anywhere within
the cell is the distance to node i from the opposing face, normalized to the cell height along that
direction. Hence ti is equal to one at node i; and zero at all other nodes as well as everywhere
on the opposing face. The transformation is given in terms of a 4x4 matrix [T], whose elements
are the 16 cofactors of the following matrix, U, made up of the cell vertex coordinates:
21
1 x1 yl z1
1 X2 Y2 Z2 (12)
1 x3 Y3 z3
I x4 Y4 z4
For example, T22 = (Y3z4-Y4Z3) +Y1 (z3-z4) +zl(y 4-Y3) The four homogeneous coordinates are
given as follows in terms of x,y,z:
tl 1
t2 [T] x (13)
t3 6V y
t4 j z
where V is the tetrahedron volume. These coordinates are especially convenient since the scalar
finite elements become functions of one coordinate only:
fi(x,y,z) = t I[T¾i +xT 2 +yTi3 +zTi4 ] (14)
VA = [.'Ti2 +YT 3 +2TT41 (15)
and the limits of integration are simplified. Most terms of (11) will reduce to integrals of prod-
ucts of two scalar functions, which have the simple result:
I 1i-ti -t 2
flfij dxdydz = 6V JdtI J dt2 I t tjdt 3 = +6(16 ij) (16)cell 0 0 0
where bij is the Kronecker delta and i and j may take on any values between I and 4.
3.5. Volume Integral Computations
The volume integral computations are carried out by visiting each cell once and adding
22
its contribution to SEE for every pair of edges that are part of the cell, excluding those edges that
are on perfect conductors. Let ij and m,n be the node indices of the endpoints of edges s and
t, respectively, I <i,j,m,n•4, i;ej, m;n. s and t are global indices, but ij,m and n are local
indices defined relative to a cell. Using the identity Vx(aVb)=aVxVb+VaxVb and noting that
the second derivatives of the linear functions f are all zero,
Vx = 2 L, x Vfj (17)
Vx(t = 2LtVfm x Vfn (18)
Considering the first term of (11) separately, note that the gradient terms (17) and (18) are con-
stants and may be taken outside the integral. Thus, cell k's contribution to the first volume
integral is
EEl 4VkSs.(k) - LSLtVf X Vfj Vfm X Vfn
4 VkLs Li4- st[ (Ti~3T; 4 -T 4 T 3 )(Zm3rn 4 -1Tm4 Zn3 ) (9
Ar(6Vk )4 (19)" (Ti4Tj2 - Ti2Tl4)(Zm4Zn2 - n 2T2n4 )
"+ (T2 - Ti372)(T72 Tr 3 -rm 3Trn2 )]
where vk is the volume of cell k. The cell's contribution to the second volume integral is
EE2 ,2S = [fifmvfj Vfn -f"f..VA -Vf (20)
-fi f.Vfj Vf,. + fjf,, Vfj - Vfm ]dYv
Again, the gradient terms are constants, so this may be evaluated using (16). The result is
23
2 4EE2 k0ErLsLg 4
S.t(k) = -20Vk • [( 1 +Uim -(I +(j,2)TiT(1)
- (1 + bid )TmT + (0 t6.)TT1 I
The submatrix SEE in (1) is the sum of SEE' and SEE2. Equations (19) and (21) are a closed-
form evaluation of the volume integral (11), written as an algebraic expression in terms of the
geometry of a cell and the constitutive parameters contained within it.
This completes the discussion of the interior finite element solution. The next two
chapters discuss the integral equations for the regions exterior to U. Those integral equations will
provide expressions for J in terms of the transverse E on the bcundary.
24
IV. Waveguide Continuity Conditions
This chapter develops the integral equations for the RF device problem illustrated in
Figure 8. These equations apply to the two waveguides entering and leaving the cavity region,
and are in the form of sums over waveguide modes. The derivation presented here is generic,
applying to any type of waveguide for which eigenmode solutions are available. Appendix B
gives specialized expressions for rectangular, circular and circular coaxial waveguides. Using
the mode sum form, the computation of matrix entries will be shown to reduce to inner products
of the mode functions with vector finite elements.
4.1. Combined-Source Integral Equation and Modal Expansion
The derivation of the integral equation uses the equivalence model shown in Figure 11.
The interior problem sees zero field outside the cavity region and equivalent electric and magnetic
currents on the open aperture. Those apertures are assumed to be planar and located in the end
walls of the two waveguides. They do not necessarily extend across the entire waveguide cross-
section (irises are allowed). The exterior problem sees zero field inside the cavity and oppositely-
directed equivalent currents. Notice that when the interior and exterior problems are superposed,
the equivalent currents cancel and the original problem is recovered. The integral equation
applies to the exterior problem: It gives the fields in the waveguides in terms of the equivalent
currents. The following approach is similar to that used by Harrington & Mautz in a moment
method solution for open-ended waveguide radiation [27).
Let the field in waveguide A (z < 0) be comprised of a unit-amplitude incident field in the
dominant mode traveling in the +z direction, plus a series of reflected modes traveling in the -z
direction. The transverse (to z) electric field may be expressed in terms of a complete set of
orthonormal mode functions gi [281:
25
-j - fA E Bt___ E-H'O EBt H O
El , H 1
:z-o z -d
(b)
-A -A EijBiH-BE ,HE H
A r(a)A
ýA A M A 6ýj-. B
E~ ~ H AB O
(c)
Figure 11. Equivalence Model for Waveguide/Cavity Problem: (a) OriginalProblem; (b) Interior Equivalent; (c) Exterior Equivalent
0o
El = e- go + E Cie•iZ (22)
i=0
The subscript t denotes transverse and the y's are the propagation constants. The dominant mode
function is go. The sum over modes i includes both TE and TM, as well as both sin(O) and
cos(4) degeneracies for circular and circular coaxial waveguides. The complex coefficients Ci
are unknowns that may be expressed in terms of the solution for the transverse fields in the
apertures, e.g.
C, = Ji E " •,ds - 60 i (23)z=O['A
where boi is the Kronecker delta. The propagation constant for mode i is related to its cutoff
wavenumber, kci, by
26
S2 (24)"Yi= kc -k
which is positive imaginary for propagating modes and positive real for evanescent modes. The
transverse magnetic field is
00
H-I = Yo)e-'Oz (t. x 0)- Ci Yi e"vi- (x xi) (25)
i=O
where 'Yj is a modal admittance:
__ -'-- (TE)
Yi = JL0 r (TM) (26)
tf-t (TEM)
The boundary condition at z=O is JAA = HixHA giving, from (25):
A - × i - + -A (27)z=0 i=0
Subst'tuting (23) gives the final form of the integral equation for waveguide A:
SYji J Et' - -d _A 2Yo0 ° (28)i=o rA z=O
Notice that the equivalent magnetic current is involved indirectly by EtA(z=O) =MA X1.
Similarly, the integral equation for waveguide B is
•, J-g I E k 'ds - Ja = 0 (29)i =0 Gn z =d
The primes signify the fact that waveguide B may be a different type or size than waveguide A,
27
so it may have different mode functions and modal admittances. The right hand side is zero
because there are no sources in wavegude B.
A general requirement that ensures uniqueness in aperture problems is that continuity of
both transverse E and transverse H must be enforced across each aperture. This usually implies
a requirement to solve for both transverse field components independently (or, alternatively, for
both M and J). However, the nature of the waveguide mode expansion links the transverse field
components through modal admittances, that is, Et and Ht are not independent. Therefore, it is
only necessary to solve for one of the two, and the obvious choice is E, for consistency with the
interior solution.
4.2. Discretization
The integral equations (28) and (29) give expressions for J that are substituted into the
boundary integral of (3). Hence, they are tested using the same trial functions, s as the interior
volume integral term. For compactness of notation, let ,j and *sBi denote the following inner
products:
Si- J . (30)rA
*i'~ =Jýs5 .ids (31)
Appendix B discusses the methods by which these integrals are computed for rectangular, circular
and circular coaxial mode functions. The testing procedure gives the equations
Esnc = 2jko0io Yo*, 0 , sEPA (32)
28
S EJ= ioi~o Yj " C r,(33)
St Si ti I I I I i
S~s = Jk 0•7oEYi sI '!4 •, s,' e'A (3
i=O
EJ = B (34)St Yi 'li *tist F
i=0
In practice, the mode sums may be truncated at 32 or less, depending on: (a) the waveguide
type; (b) the nature of the obstruction at and near the aperture; and (c) the ratio of the wave-
number to the cutoff wavenumber (more modes required near cutoff) 117]. Note that there is an
entry SE for every pair of edges s and t that share the same aperture, regardless of whether or
not they share any mesh cells. Hence, this matrix is not sparse. Ei"c has terms for all edges s
that are in aperture A.
4.3. S Parameters
The performance of passive RF devices is typically expressed in terms of their
"scattering," or "S" parameters. They may be found from those coefficients of the solution
vector E that correspond to mesh edges in the two apertures. The modal excitation coefficients
may be evaluated from these as
sE ~ACi= S si - 6i (35)EFrA
I BC, = e. (36)s E rR
The coefficient CO is the reflection coefficient, or S11. The transmission coefficient into each
mode of waveguide B is
T C Y 0 (37)
29
When there is only one propagating mode in waveguide B, rcO = S21 . If there is also only one
propagating mode in waveguide A, then the following conservation of power relationship must
hold: ISIItV+Is 2112 --=.
The derivations of this chapter, combined with those of Chapter III, provide a framework
for a computer solution for two-port RF device S parameters. The computer code implementation
and validation results are presented later in Chapter VII. The next chapter discusses how this
methodology is extended to solve for the properties of phased arrays of cavity radiators.
30
V. Periodic Radiation Condition
The cavity array problem (Figure 7) was formulated as a straightforward extension of the
RF device problem (Figure 8). It required replacing the integral equation for the second wave-
guide with one appropriate to a radiating aperture in an infinite array. This chapter gives the
form of the integral equation for an infinite periodic array and shows how it is reduced to matrix
form.
5.1. Periodic Integral Equation
The equivalence model for the cavity array problem is essentially the same as Figure 11.
However, the aperture on the right side, formerly waveguide B, is now one aperture in an infinite
uniform lattice. Therefore, the equivalent currents extend indefinitely over the outlet aperture
plane in both x and y directions.
Each radiator in the array is assumed to be excited by a unit-amplitude incident field in
the waveguide from z < 0, but the excitation phase may be different for each element in order to
produce a beam directed towards angles 0o, 0o in spherical coordinates. The phase shift as a
function of x and y is
O(xy) = e-jox e-jvYY (38)
OX = k sinOo cos4o (39)
VY = k sin0o sin4o (40)
Figure 12 illustrates the notation convention for an array lattice with an arbitrary skew angle ,y.
The aperture shape is arbitrary. The fields and equivalent currents in each aperture must have
the same magnitude as a function of x and y. A mathematical statement of the phase relationship
in (38) is Floquet's theorem 1291:
31
Y d cot'Y
Fiue1.IfiieAryo Aprurs SkwdLatc
n- ------------ - - ---- ~ d ( 1
,' dX
Figure 12. Infinite Array of Apertures, Skewed Lattice
J(x+md.+ndycoty,y+ndy) = J(x,y)e-JV,(,"a,÷dYcoty)e-j O•"ay (41)
The left side is the equivalent current in any unit cell, and the right side is the current in the unit
cell centered at the origin.
Equivalent currents 1 and M in the apertures will generate vector potentials A and F in
the half space z > d. The magnetic field due to these is
H(r) =VXA-jwF+ VV-F (42)
The integral equation results from evaluating Ht, the transverse magnetic field in the plane z=d
and using the boundary condition J=-2xHt. Since A(z=d) is entirely z-directed it does not
contribute to Ht(z=d) and the integral equation is
32
JW2X P+-'vv.Fj -J= 0 (43)
F is an integral over the source current M, and the limits of integration are infinite in x and y.
Appendix C describes how it may be transformed into the following infinite summation:
0 -J(x,y) + E E T,,mn Euc(kxmnkymn)e -kxmnxe -jkymnY (44)m=-orn=-o*
K, 2 2 1-1/2 rk -1k2 - kxmn - kymn kytnn) I (45)
= -•-x•-• kxmnkyinn (k2-k~xtn)]
,E, is the 2D Fourier transform of the transverse unit cell aperture field and k,,. and kmn are
sample points in the spatial frequency domain:
kxmn = 27m - ko sino 0cos4 0 (46)dx
kymn -n 27rmcot - kosinO0osino•° (47)d y d .
sometimes referred to as Floquet harmonics. The summation in (44) may be computed numeri-
cally because its terms decay with increasing I ml and In I, as discussed further in Section 5.3.
5.2. Discretization
The integral equation (44) is reduced to matrix form using the procedure outlined in the
previous chapter: First, solve for I and substitute it into the boundary term of (3); second,
substitute the series expansion for E; and third, substitute each ý., in turn for W. Note that the
integral equation involves the Fourier transform of the transverse aperture electric field, so its
expansion will be in terms of the Fourier transforms of the Ot's:
33
N
F, e, ,(k,,ky) (48)
t,(k,,ky) _J e'J -eJ-xYekAYdxdy (49)
where rR denotes the radiating aperture. Testing gives the matrix terms:
SS= Tmn -tmn J • e-kxmnXe-kymny ds , s,t E rR (50)m n F
where ttn denotes -t(km,kym). The limits of integration may be extended to + oo since •
is zero outside rk. Then, since 0. is a real function, the first integral may be recognized as the
complex conjugate of the Fourier transform of V,,. (The Fourier transform of a real function is
Hermitian, i.e. F(-k)=E*(k) [30:193]). Hence, a final expression for the elements of the matrix
SEJ is
EJ -s* kn9kSsJ -t(kxmnkymn) Tmn° s k n ) , s,t c- R (51)
m n
5.3. Floquet Mode Limits
The infinite sums in (51) must be truncated at some upper and lower limits _+m and 4+n.
Those limits are easily determined from the form of 4, derived in section C.4. Figure 13 is a
contour plot of J4 in dB for a typical finite element. The axes are kxhX and kyhy, where hX and
hy are the triangle (mesh cell) heights parallel to the x and y axes. This scaling ensures that the
size of the contours in Figure 13 are independent of mesh cell size. The Floquet harmonics are
superimposed as dots in the figure. From (46) and (47), their locations (for a rectangular lattice
and broadside scan) are kxhx=2rmhx/d. and kyhy=27rmh,,/d , . When the array scans away
34
15.00XOl~
/ 5 .0 0 W O O\.
0 00
1 0 .0 0 -/ /• •• .
.. . .Ix1
-15.00 -10.00 -5.00 0.00 5.00 10.00 15.00
k~hX
Figure 13. Typical Scalar Finite Element Fourier Transform (Contours in dB,sample points are Floquet modes for a square lattice with h ,, =hy = d,/5 = dy5)
from broadside, the points will move, but their spacing will not change.
A reasonable upper limit on the number of sample points kxinn and kynthat must be
included in the computation of (51) are those inside the -20 dB contour. (The product of and
ýt will be less than -40 dB for any points outside that contour.) The size of the contour is consis-
tently k~hX kyhy ±27r, but the sample spacing is inversely proportional to the unit cell
lengths d. and dy. Hence, using I k~h, , I kyhy I •! 21r in the leading terms of (46) and (47)
gives Im 1 •9 dx/hx and I n 1 d!9f"Y . In a typical problem, d, and dy are each approximately
35
.5X and the mesh cells are approximately .AX in height. In such cases limits of -5 < m,n <
5 are adequate to ensure convergence. More modes must be included when the unit cell size is
larger, or when the mesh cell size is smaller.
5.4. Active Reflection Coefficient and Element Radiation Pattern
Some of the most important output results from the phased array analysis are the active
reflection coefficient, Ra, and the active element radiation (far field) pattern. Both are functions
of scan angle. They are analogous to the reflection and transmission coefficients in the two port
RF device problem.
The expression for Ra is identical to CO in (35). Both are due to the field reflected into
the inlet waveguide, computed from the waveguide aperture field. But now those aperture fields
include the effects of a periodic radiation condition at the outlet side, and so it is the reflection
coefficient for one feed waveguide in an infinite array, i.e. the "active array reflection coeffi-
cient."
The active element pattern is analogous to a transmission coefficient. It is a measure of
the excitation strength of a plane wave (a Floquet mode) propagating away from the array.
Amitay et. al. show that the 0 and t0 polarization components of the element's far field pattern
are due entirely to the lowest order TE and TM Floquet modes, respectively 129:571. Rewriting
their expressions in terms of Fourier transforms gives:
EO secO (.fcoso + Ysino) .E*,(kxoo kvoo) (52)
= I coso -. tsin4) - "E(kxoo,ky00 ) (53)
A check for conservation of power may be made by computing the transmission coeffi-
36
cients for the two mn=O0 Floquet modes, whose admittances are Y 10 and Y2 oo for TM and TE
(see section C.5):
Te = E0 ýY 20 0/Yo = E, ý/secO/Yoo (54)
T6 = E Yj ,00oo o =/ Y = cosVY0ro-o (55)
Yo is the feed waveguide's dominant mode admittance. When the feed waveguide supports only
one mode and there are no grating lobes in visible space, the conservation of power relationship
is IRal + ITOI + IT411 = I
The derivations given in this chapter, when combined with the preceding chapters' finite
element and waveguide derivations, constitute the framework for a computer solution for the
scanning properties of cavity arrays. The implementation and validation results are discussed
later in Chapter VII.
37
VI. Periodic Boundary Conditions
The final stage in the problem formulation bridges the gap between the cavity array
problem in Figure 7 to the general array problem in Figure 4. The side walls of the cavity
"-gion (I are no longer conductors, so adjacent radiators are free to interact across those walls.
The periodic radiation condition does not account for that interaction. This chapter will show that
the general array problem may be accomplished by constructing a matrix for a single unit cell
as though it were a cavity with open-circuit side walls; then applying a mapping directly to the
matrix to enforce the periodicity condition. The necessary characteristics for a unit cell will be
discussed first, then the algorithm for the matrix mapping will be presented. Next, the side walls
will be shown to have no net contribution to the boundary functional. The specialization of the
algorithm to edges shared by the radiation boundary condition and a unit cell side wall is dis-
cussed last.
6.1. Unit Cell Representation
A typical radiator is fed by a waveguide through an aperture in a ground plane. Unlike
the cavity radiators considered in the last chapter, it has some structure projecting above the
ground plane. That structure may be enclosed by an imaginary box whose lower surface is the
ground plane at z=O. Its top surface at z=h is in free space above the radiator structure (see
Figure 4). The side walls of 0 are the unit cell boundaries. As was indicated in Figure 5, the
unit cell may be trapezoidal as well as rectangular. In fact, its shape is fairly arbitrary within
a few constraints, with some possibilities illustrated in Figure 14. The constraints are: (a) the
unit cell side walls do not cross the feed waveguide; and (b) opposing boundaries must be identi-
cal except for a translation of (d.,O) or (dycotydy). The first array (Figure 14a) has circular
waveguides whose diameter is larger than dy, so the unit cell shape along the boundaries has been
38
-- -- -- - - - - -- - -- - -- -- -- -I ------
(a) (b)
Figure 14. Unit Cell Definitions: (a) Circular Waveguide Array; (b) Rectangular Patch Array
altered. The second is a rectangular patch array, showing that the unit cell walls may cut through
the radiator's conducting structure.
The unit cell definition must be such that the Floquet condition is observed. Let fin
denote a field in the m'th column and n'th row of the lattice. Then the unit cell fields are related
by:
fm,n f 0 ,0 e-j Y
1x xd. (56)
Oy : i. dy cot-y + Oy dy
where ý, and Oy are given by (39) and (40). An example unit cell mesh is shown in Figure 15.
The two perspective viewQ show opposite sides of the mesh to illustrate the important requirement
that the surface mesh on opposing faces must be identical. Every edge on the + x or + y bound-
ary has an image edge on the -x or -y boundary respectively. Consequently, the expansion and
testing functions associated with those edges are identical except for a translation.
39
+y
+XX •X
(a) (b)
Figure 15. Example Unit Cell Mesh Showing Opposite Faces
6.2. Mapping from an Infinite System
The matrix generated by discretizing the functional represents interactions between
electric field sources associated with mesh edges. Consider the 2D triangle mesh shown in
Figure 16, which represents an infinite mesh of an infinite array. It was constructed in such a
way that opposing unit cell boundaries have identical mesh edges. Consequently, when the mesh
is replicated in each unit cell, the finite elements are perfectly aligned across unit cell boundaries.
If the finite element method were applied to the infinite problem, it would generate an
infinite size matrix. Appendix D shows formally, for a one-dimensional problem, how periodic
boundary conditions are exploited to reduce the problem to a finite matrix involving a single unit
cell. The extension to periodicity in two dimensions is straightforward. Essentially, the proce-
dure amounts to "folding" opposing unit cell boundaries onto each other. The +x and +y
boundary edges are removed from the problem and the -x and -y boundary edges will interact
with those just inside the opposing boundaries, but with an appropriate phase shift.
40
/ 577
4 5 3. ? 4 5"1 33
19212
S10 1 1
1011
8 9 6 J7 8 9 6020 1 20 ,
5 3 48 5 /18 16 17 18 168
19 2 1 014 1>,12 13 14 15 7-12 i3\
1 1 _' 10 V1! ... -, / !0 '
Figure 16. Triangular Finite Element Mesh for Infinite Two-Dimensional Periodic Problem
The first step in the procedure is to construct the matrix for an isolated unit cell, Next,
the terms involving the +x boundary are removed, and corresponding -x boundary terms are
created. Then the latter step is repeated for the +y and -y boundaries. For example, in Figure
16, edges 15 and 19 share a mesh cell (a triangle), so there will be matrix terms S15,19 and
S19,15. Edge 2 is edge 19's image, so the new entries created are S15,2=S5,1 19 exp{-jai} and
S2,15=S1 9, 15exp{joc} . Also, S19,19 will be added to S1-,,. This is necessary because of the
truncation of the mesh at the unit cell boundary. In the infinite mesh, edge 2 would interact with
itself through cells to the left and right. In the unit cell mesh, there is no cell to the left, so S2,2
is incomplete, but S19,19 is identical to the missing contribution. In the 3D problem with two-
dimensional periodicity, there will also be matrix entries tfr edge pairs that are both on the +x
or +y boundaries, representing their interaction within a cell to the left of the boundary. These
terms must also be added to their -x counterparts without a phase shift.
41
The algorithm is summarized in Figure 17. A consequence of removing mesh edges is
that some of the matrix rows and columns become entirely zero, and they must be dJeted before
the matrix solution can begin. This is consistent with the fact that unknowns associated with
image edges are not independent. Due to periodicity, once one is known, the other follows from
the periodicity condition. Unless one set of dependent unknowns is removed, the system would
be over-determined.
6.3. Boundary Functional
The algorithm discussed above only dealt with the volume integral terms from the
I. FOR EVERY EDGE s ON +x BOUNDARY:
A. LOCATE IMAGE EDGE s' ON -x BOUNDARY
B. FOR EVERY EDGE t SUCH THAT St.•0:
1. IF t IS ON THE + x BOUNDARY, THEN:
a. LOCATE IMAGE EDGE t'
b. SET S5,'t= Ss't, + Sst
c. SET S.t = 0
2. ELSE IF t IS NOT ON THE +x BOUNDARY, THEN:
a. SET St = Sst exp{jat}
b. SET S ,= St, exp{-ju.}
c. SET St =Sts = 0
II. REPEAT I FOR +y AND -y BOUNDARIES
III. COMPRESS THE MATRIX (ELIMINATE ZERO ROWS & COLUMNS)
IV. COMPRESS THE INCIDENT CURRENT VECTOR
Figure 17. Periodic Boundary Condition Algorithm
42
functional. The boundary integral must now account for the contributions from the unit cell side
walls in addition to the waveguide aperture and radiation boundary. The boundary integral may
be regarded as an expression for power flow across r. In an infinite array, the power entering
one side of the unit cell must be the same as that leaving the opposite side, so it is expected that
the side walls should have no net contribution. The following will show this to be true.
Consider the boundary integral of (3) at opposing unit cell walls parallel to the x-z plane.
The magnetic field must obey the periodicity condition, and since the outward surface normals
are opposite, the equivalent currents are
J(x+203,2) -J(x,- Y) e JQ2 (57)
= 1 dycot'Y
Any admissible trial function W (see Appendix A) must obey the conjugate relationship because
it represents waves traveling in the opposite direction, i.e.
W*(x+20,TY) = W-*(x,'-) eJ'Y (58)2
The boundary functional evaluated at the two unit cell walls is
dX
2F- fIdz I W-*(x, - 2) .J-(x, - ý-Zdx (9- j j2 *2
h2 d , +v d
FY+= fjdz f W *(X, 2).hX, 2 (60)0 d,
2
43
Using a change of variables, let T=x-2fl:
dx
h -2rY = ' W*(T+20,Lý) -J(,0ýd d61
0 2~-2
Using (59) and (60):
dXT-
hF = - r d-Idz IW-*(Tc?4).(T, -2dr (62)3 J2 '2
2
and therefore, Fy+ = -Fy... A similar procedure will show that Fx+ = -Fx , hence the unit
cell side walls have zero net contribution. A similar result has been reported for a two-dimen-
sional problem with periodicity in one dimension [161.
6.4. Radiation Boundary
A form of periodicity condition was already imposed on the radiation boundary through
the periodic integral equation. However, it did not account for the mesh truncation at the unit
cell boundary.
In order to complete the specification of periodicity conditions for edges on rR, either
of two approaches may be used: The first method is to apply the same algorithm (Figure 17)
used for the finite element boundary terms, but excluding I.B.2. The second, demonstrated by
Gedney for periodicity in one dimension 1161 used "overlap basis functions" at one boundary
edge. Those expansion and testing functions associated with points on one boundary extend into
the next unit cell. In the context of Figure 16, that amounts to: (a) removing the expansion
functions for edges on the +x and +y boundaries (16-20); and (b) extending the functions for
edges I & 2 and 3-5 into the next unit cells to the left and below, respectively. Then the periodic
44
integral equation is applied to the modified mesh. Both techniques were tested as part of the code
development and validation, giving equivalent results.
6.5. Summary
This chapter and the preceding three make up the formulation for the general phased
array problem. They are the framework for mapping the boundary value problem into a discrete
form that may be solved by computer. The matrix for the general array problem is constructed
of the three parts that were developed in Chapters III, IV and V: first, a sparse matrix due to
finite element interactions within the unit cell; second, a dense upper-left submatrix due to
waveguide interactions; and third, a dense, lower-right submatrix due to Floquet-mode interac-
tions. The side-wall periodicity conditions are implemented as a transformation of that matrix,
ah was described in this chapter. The next three chapters discuss the validation tests for three
computer codes that implement the solutions to the RF device problem; the cavity array problem;
and the general array problem.
45
VII. Validation - RF Device Problem
The goals for this simplified problem (Figure 8) were to demonstrate essential charac-
teristics of the finite element and waveguide mode implementations. The 3D vector finite ele-
ments were shown to correctly predict the field behavior at conductor edges and dielectric inter-
faces. The higher-order waveguide mode calculations were validated through comparisons with
measured and published results for several waveguide discontinuity problems. These results also
provided estimates for sampling requirements: the number of finite elements; and the number
of waveguide modes. More complete details are given in a previous report [171.
7 1. Computer Code Implementation
7.1.1. General Procedure. A FORTRAN computer code named TWOPORT implements
the solution to this generic problem. An outline of the actions it takes is given in Figure 18. The
user instructions, read during the first step, include: the type, size, location, number of modes
and Er for each waveguide; the frequency limits and frequency stepsize; and the name of the file
containing the problem geometry. The geometry file contains three sections: The first lists the
node coordinates and several flags for each, identifying boundary nodes (port # for those in
waveguide apertures and conductor # for those on conducting surfaces). The second block lists
the indices of the four nodes comprising each tetrahedron and the index of the material filling it.
The last block lists the complex Er and tr of each material.
The third step converts the node-based geometry to an edge-based geometry. Each edge
in the mesh defines an electric field vector expansion function that exists over all cells adjacent
to that edge. If two nodes are on the same conductor, then there cannot be a field along the line
joining them, so those edges are not included in the edge list. This is the means of enforcing the
boundary condition on tangential electric field at conducting surfaces. On the other hand, if two
46
I. READ INSTRUCTIONS AND OPTIONS
II. READ PROBLEM GEOMETRY
III. CREATE EDGE-BASED GEOMETRY
IV. FOR EACH FREQUENCY:
A. COMPUTE TERMS OF SEE ACCORDING TO (20), (22)
B. FOR WAVEGUIDE A:
1. COMPUTE INCIDENT CURRENT VECTOR IN (32)
2. COMPUTE TERMS OF SF FROM (34), (35)
C. FOR WAVEGUIDE B, COMPUTE TERMS OF SEJ FROM (34), (35)
D. SOLVE (SEE + SU ) E = Elia FOR E
E. COMPUTE MODE EXCITATION COEFFICIENTS FROM (36), (37)
Figure 18. Solution Procedure in Program TWOPORT
nodes are on different conductors, there may be a field between them, and such edges must still
be included. The matrix fill operations must account for the direction of the vector function, so,
by convention, it is always directed from lowest to highest node index.
Most of the actions under step IV in Figure 18 are straightforward implementations of
formulas derived in Chapters III and IV and Appendix B. The matrix elements are computed in
three steps: first, the interior finite element interactions; and second and third, the exterior
waveguide interactions for waveguides A and B, respectively. After solving the system for the
unknown electric field coefficients, excitation coefficients for any number of higher order modes
may be computed.
7.1.2. Matrix Solution. The matrix structure is mostly sparse, except for two dense
submatrices in the upper left and lower right corners. This structure results from ordering the
47
edges by increasing centroid z coordinate, starting with those in the waveguide A aperture and
ending with those in the waveguide B aperture. Unfortunately, there is not usually any special
structure (sparsity pattern) to the matrix. It tends to be banded, but the bandwidth is large and
there is no advantage to using band storage or specialized band solvers. Hence the two approach-
es used for the matrix solution were: (a) ordinary LU decomposition (LUD) using standard
library routines and ordinary row-column storage (IMSL 131:341 and LAPACK[32:150]) for
problems with 2000 unknowns or less; and (b) the conjugate gradient method (CGM) based on
formulas from Sarkar and Arvas 1331, using sparse storage, for larger problems.
The CGM solver was written specifically to accommodate the form of sparse storage most
attractive for this particular class of problems. The two dense submatrices due to the waveguides
are stored in ordinary row, column format; while the sparse finite element matrix is stored in a
column array. Its entries are in arbitrary order, stored in the order that each edge pair is first
encountered as the fill algorithm performs operations one cell at a time. That assembly technique
is considerably more efficient than performing the same operations one edge at a time [261.
Figure 19 is a comparison of execution times versus number of edges for LUD and
CGM. The CGM solver clearly has the advantage for problems with 1000 unknowns or more.
Unfortunately, its solution time is more difficult to predict a priori, since the number of iterations
it will need to converge is unknown. The convergence measure is the residual error norm, which
is the L2 norm of the solution error:
e= Ii S i- E 14c (63)
(i is the iteration number). The initial guess Eo is the zero vector. Figure 20 is an example
convergence history (for a microstrip transmission line fed by coaxial waveguide at each end).
The S parameters have converged to within 1 % of their correct values when El/E0 is less than
.001. Problems with rectangular and circular waveguide ports typically only require I N-2N itera-
48
1.oE.06
S1.02E054)
.O. .E-04 : . :Wk-- 1.0E+03
0
0.
.- I 1.0E÷02 :•"
o) " ED LUD ' CGM
1.02E01100 1000 10000 100000
NUMBER OF UNKNOWNS
Figure 19. Comparison of Solution Times for LU Decomposition (IMSL) and ConjugateGradient Method (CPU seconds on VAX® 8650 minicomputer)
tions to converge, where N is the number of unknowns.
7.2. Waveguide Discontinuities
7.2.1. Iris in Coaxial Waveguide. Two test cases are representative of the several
waveguide discontinuity problems discussed in [171: a conducting iris in coaxial waveguide; and
a step discontinuity in circular waveguide. For the first of these, Figure 21 is the tetrahedron
mesh used as the input to program TWOPORT. Relating this model to Figure 8, the inlet and
outlet waveguides are both coaxial (inner radius a= 1.5mm, outer radius b=3.5mm) and the
cavity region 0 represented by the mesh is simply a short section of the same waveguide.
Shading has been added to identify the nodes tagged as perfect conductors. One quadrant of the
inlet end is blocked by a thin conducting iris.
Measurements of this "device" were made by inserting a foil iris between two APC-7mm
adapters and using a network analyzer to obtain S1 over a 2-18 GHz frequency range. Figure
49
1.OE-0 1
IL
z 1.OE-02 0.-0
-JJ
LLW
1.OE-0400 1 2 3 4 5 8 7 8 9 10
ITERATIONS/UNKNOWNS
Figure 20. Convergence of Residual Norm and Reflection Coefficientusing Conjugate Gradient Matrix Solver
Figure 21. Finite Element Mesh for Coaxial Waveguide Section with Conducting Iris
50
22 is a comparison of those measurements with TWOPORT calculations, for both the magnitude
and phase. These results validate two important features of the solution approach: First, the
higher-order coaxial mode functions and their inner products with finite elements are correctly
implemented. Second, the finite element solution is correctly predicting the field behavior at the
edge of a perfect conductor.
Z2.2. Circular Waveguide Mode Converter. A step discontinuity in a circular wave-
guide is a simple type of mode converter commonly used in feeds for reflector antennas 1341.
The antenna's radiation pattern shape is controlled by carefully adjusting the amplitude ratio of
modes in an oversize (multimode) waveguide or horn. Figure 23 shows finite element meshes
for two test cases with different inlet waveguide radius and the same outlet waveguide radius.
Multimode calculations for these geometries were presented by Masterman & Clarricoats [35].
Their results are in terms of a mode conversion ratio, M:
W 1.2 180"o'10
APC-7mm CoaxS/F•_ \ 165 U'
•-0.8- 0 "• •150 I-Z 1Z08 FOIL IRIS Z
,FL 0.6- 0 -135 L"
u- LW W0 __ __ _ 00 0.4 - Meas Magnitude + ÷120 0Z Z0 Meas Phase 0
0 0.2 - HFEM MagnitudeI-. HFEM Phase u.
cc 0 1, 90 "
2 4 6 8 10 12 14 16 18
FREQUENCY (GHz)
Figure 22. Measured and Computed S1I Magnitude and Phase for Coaxial Iris
51
M Cgp2(P =a) (64)
SCigp I(p =a) I
C' and Cj are the excitation coefficients for the TE, 1 and TM modes, respectively, and g', and
g92 are the radial components of the mode functions. Hence, M is the relative strength of the
two modes measured at the wall of the outlet waveguide.
Figure 24 compares TWOPORT calculations with the multimode results. The disconti-
nuity at 6.25 GHz for a, = 1.15" is due to the fact that the inlet waveguide also supports the TMI
mode above that frequency. The agreement of the TWOPORT calculations with the well-estab-
lished multimode calculations indicates that the higher-order circular waveguide mode functions
and their finite element inner products have been implemented correctly. It also demonstrates
a flexibility of the approach of using higher-order modes in modeling feed structures that support
more than one propagating mode.
7.3. Printed Circuit Devices
One of the most important capabilities of the finite element method is its ability to deal
with inhomogeneous dielectrics. Printed circuit devices are an example application that is espe-
cially interesting to this research because their modeling requirements are similar to the printed
antennas discussed in the introduction. The specific test cases were: (a) a straight length of
microstripline with coaxial connectors; and (b) a microstrip meander line.
7.3.1. Microstrip Transmission Line. The finite element mesh shown in Figure 25 is a
thin dielectric slab (100 pm), situated in the bottom of a perfectly conducting box. The mesh for
the air space above the slab is not shown. Shading has been added to identify the nodes associat-
ed with microstrip line, the coax center conductor, and the coax dielectric. There is an identical
coaxial port at the far end of the device. The coax dimensions (a=43 pm, b= 100 pm) were
52
r a, (a) ýa 1 (b)
Figure 23. Finite Element Meshes for Circular Waveguide Step Discontinuities(Mode Converters) with a2 =1. 4": (a) a1 1. 05 "; (b) a1 1. 15.
-27
S-8
9) -1+HFM .0
w> 1 Multimode, a, -1.05'
0 z-2 HFEM, a, -1 .050
- aI I
5.2 5.4 5.6 5.8 6 6.2 6.4 6.6 6.8FREQUENCY (GHz)
Figure 24. Mode Conversion Ratio for Circular Waveguide Mode Converters:TWOPORT and Multimode 135] Calculations
53
CASE OUTUNE
MICROSTRIPi UNE
, 400 Irn
COAX APERTURE
Figure 25. Microstrip Transmission Line Section in Conducting Enclosure
chosen strictly for convenience since the objective of this test was to verify a microstrip
transmission line property (guide wavelength).
The substrate dielectric constant was chosen as 12.9 to represent Gallium Arsenide
(GaAs). The 75/Lm-wide microstrip line has the same characteristic 50 1 impedance as the coax.
The enclosure dimensions were chosen to ensure that there are at least 4 line-widths space
between the microstrip and the cavity side and top walls. That is adequate to ensure that the
enclosure does not influence the guide wavelength or characteristic impedance 1361.
The transmission coefficient for a line length of 500 Am was computed first. Next, the
calculation was repeated with finite element geometry scaled by 1.5, giving a line length of 750
Am. The difference in S21 phase was 350 at the 40 GHz test frequency, from which the guide
wavelength is calculated as 2.57 mm. Quasi-static formulas developed by Wheeler 137] give 2.54
mm. This close agreement indicates that the finite element method is correctly predicting the
fields at the interface between highly contrasting dielectrics, even in the presence of a sharp
conducting edge.
54
7.3.2. Microstrip Meander Line. Since the modeling method accurately accounts for all
relevant boundary conditions for a simple microstrip line, it should be obvious that it can correct-
ly predict the performance of most passive printed-circuit RF components. A microstrip meander
line was chosen as a demonstration case since measured data was available. The circuit dimen-
sions (in Arm) and the finite element mesh (substrate cells only) are shown in Figure 26.
The measurement reference planes are at the centers of the pads at each end of the circuit,
while those used in the calculations are at the points where the lines begin to taper from 75/Lm
down to 251im, so the difference must be accounted for when comparing the data. The circuit
layout shows the location of via holes that provide a ground for the measurement probes. At the
contact points, the 7 5gum center conductor and 50jm-wide slots form a 500l coplanar waveguide
(CPW). Using formulas given by Rowe and Lao [381, the effective relative permittivity for that
transmission line structure was found as feff=6. 29 . The measured S, , phase was corrected by
adding 3600°. (75pm)(Eefr.f)4/Xo. The calculated S21 phase was also corrected by subtracting the
excess phase introduced by the coaxial connectors, computed from a straight length of microstrip
line as discussed in the preceding section. Figure 27 compares the measurements and calculations
over the frequency range from 1 to 26 GHz. The slight discrepancy in the phase slope is
attributable to the differing reactances of CPW-microstrip (measurement) and coax-microstrip
(calculations) transitions.
7. 4. Importance of Higher Order Modes
An important issue is whether there is in fact any benefit to ising higher-order modes in
conjunction with a finite element solution. The alternative is to extend the finite element mesh
into the connecting waveguides fE.r enough that any higher modes excited by the cavity's contents
and/or apertures have decayed to insignificance 1391. The answer depends on two factors: first.
the mode excitation strengths; and second, their attenuation cons•ants in the waveguides, which
55
25
05
2 0) 2
w i.. -30
-40-
-0
4 -0 -
Measured (raw data)Zi-8 - Measured (corrected)
- 90 - Calculated
-100- ----
0 5 10 15 20 25 30FREQUENCY (GHz)
Figure 27. Comparison of Measured and Calculated S., Phase forMicrostrip Meander Line
56
depends mainly on the ratio of the frequency to the mode cutoff frequency.
Figure 28 shows how the magnitude of S11 converges with the number of modes in the
one-quadrant coaxial iris problem (Figures 21 and 22). The most important modes are the TEM
and the TE1 1 modes. The latter is excited at about -10 dB, and its cutoff frequency in APC-7
coax is about 20 GHz. At 18 GHz it decays by 29 dB per wavelength, so in order for it to decay
below -30 dB (regarded as negligible for purposes of S parameter calculation), the mesh would
need to extend roughly /%X into the waveguide in each direction. The finite element mesh and
the interaction matrix would then contain an additional 500-1500 edges and unknowns,
respectively.
On the other hand, the two microstrip problems do not excite any higher order modes
above the -30 dB level, so those devices could be accurately modeled using only the lowest-order
mode and still terminating the mesh at the waveguide aperture. Thus, the use of the mode sum
as a continuity condition can make the finite element solution more efficient, but the improvement
WU 1APC-7m, Coax1
D
S0.8 0 .^
0 FOIL IRIS
t 0.6-_ U
LL '
o 0.4 -
0 1 Mode U,- 0.2 1 Mo
C' 01-11 Modest_-- O-L01-21 Modes I -- J ,
Cr 02 4 6 8 10 12 14 16 18
FREQUENCY (GHz)
Figure 28. Reflection Coefficient for One-Quadrant Coaxial Waveguide IrisCalculated with Varying Numbers of Higher-Order Modes
57
over the dominant-mode-only approach is problem-dependent.
7.5. Summary
In addition to validating the essential properties of the electromagnetic solution approach,
the TWOPORT code provided valuable data on practicality: typical mesh sizes and execution
times. The data, summarized in Table I, includes all of the test cases discussed in this chapter,
plus three others: The rectangular and circular waveguide cases were short sections of waveguide
used to verify dominant mode propagation; and the coax-to-rectangular waveguide transition
consists of a right-angle metal launcher in a section of rectangular waveguide approximately X/3
long. The meshes for these three problems are shown in Chapter VIII, where they are used as
cavity array elements.
Table I. Mesh Sizes, Iterations and Matrix Solve Time for TWOPORT Test Cases(Solve time is CPU time on VAX® 8650 minicomputer)
SolveTest Case Volume/X3 Number Cells/X 3 Number of Time
Mesh Cells Iterations (min.)
Rectangular Waveguide .149 348 2067 .9N 0.7
Circular Waveguide .120 342 2850 1.3N 1.4
Iris in Coaxial Waveguide .0135 847 63x103 .9N 12.5
Circular Waveguide .376 1504 4000 0.5N 43.5Mode Converter
Coax-to-Rectangular .294 2442 8294 4.7N 226.Waveguide Transition
Microstrip Meander Line 9.4x10-5 4190 4.5x107 8.4N 524.
Microstrip Transmission Line 5.7x10-4 4662 8.2x 106 7. IN 683.
The first column in Table I gives the total volume in cubic wavelengths for each problem.
For the simpler problems, the number of cells is determined primarily by the sampling
58
requirement of 10 nodes per wavelength. More complicated structures, however, require more
cells in order to capture fine details of geometry. Column 4 gives the number of iterations in
terms of N, the number of mesh edges, that were required to obtain Lonvergence of the residual
norm to .001 of its initial value (using the zero vector as an inital guess). In all cases, the
maximum number of iterations was less than 10 times the number of edges (unknowns). The
matrix solve time per frequency sample is given in the last column. These results indicate that
practical design problems can be accomplished on typical minicomputers and workstations.
This chapter has validated two key elements of the solution approach. First, it showed
that FEM can correctly account for boundary conditions typical of antenna problems: conductors,
conductor edges and dielectric interfaces. Second, it showed the validity and usefulness of the
waveguide mode integral equation as a continuity condition. These results validated not only the
generic concepts, but the computer code implementation as well. The latter was especially
important since the TWOPORT code included most of the structure and modules needed for the
subsequent cavity array problem solution discussed in the next chapter.
59
VIII. Validation - Cavity Array Problem
The second development stage replaced the integral equation for the outlet waveguide
with the periodic integral equation. This allows validation of the periodic radiation condition
without the complexity of side-wall periodicity conditions. It also provides a tool useful for
analyzing the properties of radiators such as open-ended waveguides, cavity-backed slots and
multimode horns. This chapter describes its implementation in a computer code and presents
comparisons of its calculations with published results. Also presented are two specific examples
of radiators that are beyond the capability of previous methods: a pyramidal horn; and a rectan-
gular waveguide with a coaxial transition in close proximity to the aperture.
8.1. Computer Code Implementation
A FORTRAN program named CAVIARR (cavity array) implements the solution to this
generic problem, depicted in Figure 7. Figure 29 is an outline of its actions. Note that all of
these actions up to step IV.C. are essentially the same as in program TWOPORT. The interior
finite element matrix and the inlet waveguide submatrix calculations are exactly the same, since
they do not depend on the scan angle. For each separate scan angle, the submatrix due to the
radiating aperture must be recalculated, then the system must be solved again for the field vector
E. The active reflection coefficient calculation is the same as the calculation for S1 l in TWO-
PORT. However, instead of the S21 calculation, CAVIARR calculates the element far field and
transmission coefficients into 0 and 0 polarizations.
8.2. Waveguide Arrays
8.2.1. Rectangular Array. The scan-dependent reflection coefficients for several open-
ended waveguide arrays are available from publications by other authors. For example, the
properties of the rectangular waveguide array shown in Figure 30a were first presented by Dia-
60
I. READ INSTRUCTIONS AND OPTIONS
II. READ PROBLEM GEOMETRY
III. CREATE EDGE-BASED GEOMETRY
IV. FOR EACH FREQUENCY:
A. COMPUTE TERMS OF SEE ACCORDING TO (20), (22)
B. FOR WAVEGUIDE A:
1. COMPUTE INCIDENT CURRENT VECTOR IN (32)
2. COMPUTE TERMS OF SE FROM (34), (35)
C. FOR EACH ANGLE:
I. COMPUTE TERMS OF SEJ FROM (51)
2. SOLVE (SEE + SEJ ) E = Einc FOR E
3. COMPUTE REFLECTION COEFFICIENT AND MODEEXCITATION COEFFICIENTS FROM (36)
4. COMPUTE ELEMENT FAR FIELD AND TRANSMISSIONCOEFFICIENTS FROM (52)-(55)
Figure 29. Solution Procedure in Program CAVIARR
mond [401 and independently confirmed by several others [291,[411.
Figure 30b is the finite element model used by program CAVIARR - simply a short
section of waveguide. The nodes on the shaded surfaces were tagged as conductors, while those
on the front and back faces were tagged as radiating aperture and waveguide boundaries, respec-
tively.
Figure 31 compares Diamond's results with CAVIARR computations. The interesting
feature of this test case is the scan blindness near 330 in the H-plane (the 0=0 plane). When the
array scans to that angle, nearly all of the incident power is reflected back into the waveguide.
61
b... . . ...40.
.. ...Y. . (b )
Figure ~ ....... 30 Retn.arW...d.etCae.a.rryLticad(b erhernM s
0 .. .. . . . ... ... .. ...0 . . . . . . ..
-4 . .. .. . .. ... .. ... -8. .. ..
........z. ......-1 2. .. .. . ... .. . .. . . . . .
b .4102304500
Figure 0. Rectngular Waveguide Test Case: a ra atc;ad()TtaernMs
06
The CAVIARR (HFEM) calculations used waveguide modes up to m,n=2,3 and Floquet modes
up to I ml, In 1 =5,5. This number is consistent with the estimate given earlier in Section 5.3
since the mesh spacing in the aperture is approximately V110. In fact, I ml, In I <-4 was adequate
to ensure convergence of the active reflection coefficient to within 1% in magnitude and 1V in
phase.
8.2.2. Rectangular Array with Conducting Iris. A straightforward method for eliminat-
ing the H-plane scan blindness in the rectangular waveguide array is to place conducting irises
in the apertures. For example, Lee & Jones 411] performed a multimodet analysis for the same
array lattice and waveguide size as in Figure 30, except that a thin conductor blocked the left and
right 1h of each aperture. The same finite element mesh as before (Figure 30b) was used for this
problem, except that the nodes associated with the iris were tagged as separate conductors.
Figure 32 compares the CAVIARR calculations with the H-plane element gain pattern from [411.
The scan blindness was indeed suppressed by the addition of the irises, but at the expense of a
reduction in the broadside gain.
The cdose agreement between the CAVIARR calculations and published results for these
two test cases demonstrates that the hybrid periodic integral equation/finite element formulation
is valid. It also demonstrates that its implementation in the computer code is accurate.
8.2.3. Circular Waveguide Arrays. The lattice geometry for an array of circular wave-
guide radiators is shown in Figure 33a. The incident waveguide field is in the dominant TEII
mode and polarized parallel to the y axis. Multimode calculations for this geometry are available
from Amitay et. al. [291.
1 The multimode method is sometimes referred to as a moment method. It equates the transverse fields
on the two sides of the aperture as sums over modes appropriate to each region. The modes of one waveguideare used as testing functions in order to generate a matrix equation. The generic technique developed for wave-guide discontinuities was adapted to the phased array problem using Floquet modes.
63
-4-
z0W -12
u -16"cc
No Iris /
-20 - Multimode, 50% Iris
21 HFEM, 50% Iris
-2410 10 20 30 40 50 60
SCAN ANGLE 0 (deg), H-PLANE
Figure 32. Active Reflection Coefficient vs. Scan Angle for Rectangular WaveguidePhased Array with Conducting Irises
. . . . . ..:: .:+ . :. :. -. . :. . :. :
(a) (b)
Figure 33. Circular Waveguide Array Test Case: (a) Lattice; and (b) Tetrahedron Mesh
64
-~~~ . . .. . . . .. . . .ilii
D Multinode, E-Plane
* HFEM, E-PlaneZ 0.8-L 0.Multimode, H-Plane
UHFEM, H-Plane
Z 0.6 UW
U-
0 1400
-JL-w
Cc 00 10 20 30 40 50 60
SCAN ANGLE 0 (deg.)
Figure 34. Circular Waveguide Array Active Reflection Coefficient vs. Scan Angle
The finite element mesh used for CAVIARR is shown in Figure 33b. Its radius was
adjusted so that the sum of the tetrahedron volumes was the same as a perfect circular cylinder
with .343Xo radius. Figure 34 shows the finite element calculations along with the published
multimode results for both E- and H-plane scanning [29:276,2801. Again, the CAVIARR code
is accurately predicting the active reflection coefficient at all scan angles.
A method for suppressing the scan blindness in circular waveguide arrays is to use dielec-
tric-loaded waveguides that can be packed closer together since their radius is smaller for a given
wavelength. In the following test case, the waveguide radius was adjusted so that Er'/h a =
.343Xo. The lattice spacing was adjusted so that dx/a and dy/a are constant. The finite element
model was simply a scaled version of the mesh in Figure 33b. Figure 35 is a comparison of the
loaded and unloaded cases, the latter with Er= 4 . 1, for H-plane scanning 129:2901.
A third and final test case involving circular waveguide arrays also includes a dielectric,
65
LU0.. - Multimode, Loaded
* HFEM, LoadedZ 0.8O Multimode, No Diel.
z 0.6
LL ~ * ----LUo 0.40
z0-0.2
-JU.
0 I0
0 10 20 30 40 50 60
SCAN ANGLE e (deg.), H-PLANE
Figure 35. Active Reflection Coefficient vs. Scan Angle - Circular WavegiideArray with and without Dielectric Loading
but instead of completely filling the waveguide, it is only a short plug in the aperture end, flush
with the ground plane. Amitay et. al. give results for E-plane scanning using the lattice in Figure
33a, and .429Xo length plugs with fr= 1.8 [29:293]. The finite element model for this problem
was a longer version of that depicted in Figure 33b - it was 6 mesh cells in length, in order to
meet the X/10 sampling requirement. Figure 36 compares those results for the no-dielectric case.
The results of the CAVIARR and multimode computations are substantially identical. (There are
sources of error in both approaches, such as the number of modes and the number of integration
points. The important fact is that both methods are accurate enough to allow a judgement of
whether or not an antenna design is acceptable.) This demonstrates the important capability of
the hybrid finite element method for modeling arrays that have dielectric loading. Unlike the
multimode method, it is not restricted to homogeneous dielectrics.
66
1
a Amitay, No Diel. .4ET.429 ),o
*- HFEM, No D0el. H ro 1.8z 0.8O - Amitay, Dlel. Plug
0 HFEM, Diel. Plug
"Z 0.6
U-;-1U./W
U' 0 .4 m 0 0 3 200z X
'- 0.20
U-U.ir 0'
0 10 20 30 40 50 60
SCAN ANGLE ()(deg.), E-PLANE
Figure 36. Active Reflection Coefficient vs. Scan Angie - Circular WaveguideArray with and without Dielectric Plugs
8.3. Pyramidal Horn Array
The multimode method was the only rigorous technique available for addressing cavity
array problems. This confined the available solutions to structures for which mode sets can be
constructed, mostly cylindrical waveguides. There is, however, one example by Amitay & Gans
[421 that uses a variation of the multimode method to approximate the scanning characteristics
of an array of pyramidal horns. Their technique models the pyramidal horn, shown in Figure
37a, as a rectangular waveguide whose dimensions are the same as the horn mouth, containing
planes at various z locations that are transparent to those waveguide modes that may propagate,
and shorting those that are cut off. In spite of the approximate nature of their prediction tech-
nique, they obtained fairly good agreement with measured data.
This problem is a fairly stressing case for the hybrid finite element method. First of all,
the mesh, depicted in Figure 37b, has approximately 6000 edges. Furthermore, the large unit
67
a= .9051
dx= 2.78X
(a) (b)
Figure 37. Pyramidal Horn Radiator: (a) Dimensions; (b) Tetrahedron Mesh
cell size (the same as the horn mouth) requires an unusually large number of Floquet modes (60
in kx and 15 in ky).
Figure 38 is a comparison of the CAVIARR calculations for active element gain with
measured data from [42]. The scan plane is 0=90°, for which the co-polarized field is the E0
component. The gain in decibels is 20 logjoEO. The interesting feature of this test case is the
scan blindness near 400, due to excitation of a higher-order waveguide mode at the aperture. In
contrast to the waveguide cases discussed earlier, the active reflection coerficient does not become
large near the blindness angle. Since there is more than one propagating Floquet mode in this
instance, incident power that is not transmitted in the desired direction is transmitted into another
mode, i.e. a grating lobe.
68
8.4. Coaxial-to-Rectangular Waveguide Launcher
An end-wall transition, or "launcher," from coaxial to rectangular waveguide is shown
in Figure 39. This type of transition ha.z an advantage in a rectangular waveguide phased array
because the waveguides are packed too closely to use a broad-wall launcher. This design, due
to Tang and Wong [431, is an extension of the coax center conductor with a shorting post joining
it to the broad wall. The probe is centered in height and offset 1/10th the waveguide width. Its
length should be approximately one fourth the guide wavelength at the design frequency. For
a length of 9mm, the frequency response is shown in Figure 40 (computed by program TWO-
PORT).
The experimental array reported in 143] had a long waveguide section between the launch-
ers and the open apertures, and higher-order waveguide modes excited by either one would not
affect the other. The antenna's weight and size would both benefit if the launchers were placed
as close as possible to the apertures, but then their mutual interactions are not negligible.
z -1 0 ! a
w -15--
J -20-
SMeasured-25 '.•
"Calculated (HFEM)
-300 10 20 30 40 50 60
E PLANE SCAN A11GLE (deg)
Figure 38. Active Element Gain vs. E-Plane Sc m Angle tb(r PyramidalHorn Array (Measured Data tronm Amitav & (11nm 1421)
60)
4a
ay
(a)(b
Figure 39. End-Wall Transition from Coax to Rectangular Waveguide (After Tang &Wong [431): (a) Geometry; (b) Cutaway of Tetrahedron Mesh
wFRON T VIEW SIDE VIEW
_9.2mm EEI
0.-4.66 _ n< -
z!J 0.6 -
U-
0 0.4
z0U 0.2
LA-
7 7.5 8 8.5 9 9.5 10 10.5 11 11.5 12
FREQUENCY (GHz)
Figure 40. Reflection Coefficient (SI I) for Coax - Rectangular Launcher:Probe Length =9mm, Post Height =5mm, X-Ban,) Waveguide
70
Considering again the rectangular waveguide test case discussed in Section 8.2.1, suppose the
waveguide elements are only 10mm deep, with a launcher extending to within 2.5mm of the
aperture. The predicted active reflection coefficient vs. angle is shown in Figure 41, compared
to the original case, where the array is fed by semi-infinite rectangular waveguides (both calcula-
tions by CAVIARR). Evidently, the launcher in close proximity to the waveguide opening
prevents the formation of the higher-order waveguide mode that is responsible for the scan
blindness condition, which was the purpose of the conducting irises discussed by Lee & Jones.
Furthermore, the interactions between the launcher and the waveguide aperture do not generate
any additional resonance effects, so there is no need to separate the two by a long length of
waveguide. Thus, the array element consisting of a coaxial launcher in a short waveguide section
is a better solution than was previously available because it achieves the same result with a
smaller and simpler structure. Although rectangular waveguide arrays are outdated, this is
nonetheless an illustration that: (a) the hybrid finite element method can be used to solve practical
>-4-z
-12
-20 - Infinite Waveguide I
-24 -- i7T-0 10 20 30 40 50 60
SCAN ANGLE 0 (deg), H-PLANE
Figure 41. Active Reflection Coefficient for Rectangular Waveguide Array withCoaxial Launcher and Rectangular Waveguide Feeds
71
problems for which accurate methods were not available before; and (b) the cavity array solution
has important applications by itself even though it, like TWOPORT, was an intermediate step in
the code development. The following chapter describes the implementation and results from the
final step, which implements the periodicity conditions at unit cell side walls for general array
radiators.
72
IX. Validation - General Array Problem
The third and final stage of computer code development and validation implemented the
periodicity condition for unit cell side walls, removing the requirement that the radiators be
separated by conducting walls. Initial validations were performed for open waveguide arrays
using the previous chapter's results. Test cases for arrays of microstrip patches and monopole
arrays were used for further validation, the latter including a hardware experiment. Last, perfor-
mance predictions for flared notch and printed dipole arrays are presented to demonstrate the
method's versatility.
9.1. Computer Code Implementation
A FORTRAN program named PARANA (phased array antenna analysis) implements the
general array problem solution. Figure 42 is an outline of the program's actions. Up to step
IV.C. it is essentially identical to CAVIARR, except that a listing of image edges is created in
step III. These identify the -x and -y boundary counterparts tfr each edge on the +x and +y
unit cell walls. This places an additional requirement on the geometry file - the unit cell bound-
ary nodes must be tagged so that these edges can be identified. In addition, it requires the mesh
generator to create the grid in such a way that the surface mesh on opposing faces is identical.
Although the CAD software (I-DEASTM [441) does not have any provision for enforcing this
requirement, it results naturally when the mesh areas and constituent curves are defined in
consistent and logical order.
In step IV.C. I., the calculation for the radiation boundary terms of SE is slightly modi-
fied to include special handling for -x and -y boundary edges, as discussed in Section 6.4. Step
IV.C.2. is a straightforward implementation of the algorithm shown in Figure 17. One of its
effects is to zero out the rows and columns of SEE corresponding to +x and +y boundary edges.
73
I. READ INSTRUCTIONS AND OPTIONS
II. READ PROBLEM GEOMETRY
III. CREATE EDGE-BASED GEOMETRY AND LIST OF IMAGE EDGES
IV. FOR EACH FREQUENCY:
A. COMPUTE TERMS OF SEE ACCORDING TO (20), (22)
B. FOR WAVEGUIDE A:
1. COMPUTE INCIDENT CURRENT VECTOR FROM (33)
2. COMPUTE TERMS OF SE! FROM (34)
C. FOR EACH ANGLE:
1. COMPUTE TERMS OF SEJ FROM (51) USING OVERLAP ELEMENTS
2. IMPOSE PERIODICITY CONDITION ON SEE ACCORDING TO FIG. 17
3. ELIMINATE ZERO ROWS & COLUMNS FROM (SEE + SEJ
4. SOLVE (SEE + SFJ ) E = Einc FOR E
5. RESTORE BOUNDARY EDGES
6. COMPUTE REFLECTION COEFFICIENT AND MODEEXCITATION COEFFICIENTS FROM (36)
7. COMPUTE ELEMENT FAR FIELD AND TRANSMISSIONCOEFFICIENTS FROM (52)-(55)
Figure 42. Solution Procedure in Program PARANA
Step IV.C. 1. had a similar effect on SEJ. Before starting the matrix solution, the matrix is
compressed to eliminate zero rows and columns. After the matrix is solved, the periodicity
conditions are used to solve for the field along the +x and +y boundary edges.
9.2. Waveguide Arrays
The PARANA code is also capable of modeling open-ended waveguide arrays, although
74
less efficiently than CAVIARR. A comparison between the two was a means for verifying the
periodic boundary condition while keeping all other details of the problem and solution the same.
Figure 43 is the tetrahedron mesh used as the input to PARANA. It is a shallow slice of the unit
cell in free space outside the waveguide, above the ground plane. The shaded area identifies
those nodes and cells bordering on the ground plane. Contrast this to the CAVIARR geometry
model (Figure 33b), which is a slice of waveguide below the ground plane.
The test case parameters were: a = .3Xo , d, = .88Xo ; dy = .733Xo and T = 600
(equilateral lattice). Figure 44 compares the computed results from the two codes for scanning
in the H-plane (x-z plane). The fact that they are essentially identical is evidence that the periodic
boundary condition is working as proposed. The results for E-plane scanning were similar, with
the same degree of agreement between the two codes.
A second waveguide test case was a rectangular lattice of rectangular waveguides. The
lattice dimensions were the same as the waveguide size, i.e.: a = d, = 23mm and b = d -=
10mm. The finite element mesh was similar to Figure 30 except that there were three layers of
cells, each with the same thickness. The perimeter edges of the first layer were tagged as con-
z
ZX4--
z
Figure 43. Finite Element Mesh for a Skewed-Lattice, Circular Waveguide Array Unit Cell
75
1• •Cavity Array Code
- 8 General Array CodeZ 0.8-
u0.6
I.LUo 0.4
0 04
Z0'-, 2 *'
-J *IL
0 10 20 30 40 50 60SCAN ANGLE 0 (deg.), H PLANE
Figure 44. Circular Waveguide Array Active Reflection Coefficient - Comparison of ResultsUsing Cavity Array (CAVIARR) and General Array (PARANA) Models
ductors to indicate that the mesh extends part way into the waveguide. The perimeter edges of
the other two layers were tagged as unit cell walls. The calculations by PARANA for 10 GHz,
shown in Figure 45, are again essentially the same as the CAVIARR results. This test case
illustrates some potentially important flexibilities of the implementation: the mesh may extend
into the feed waveguide; and the feed waveguide may adjoin the unit cell boundary.
9.3. Microstrip Patch Array
The first of four demonstration cases uses a simple rectangular microstrip patch on a
substrate that is thick enough to support a surface wave in the dielectric. The geometry, shown
in Figure 46, has the dimensions (in wavelengths) used by Pozar in MoM calculations [451. The
substrate has relative permittivity of 12.8 to represent GaAs. Since the input impedance of the
patch is very low (less than 5fl) due to the substrate thickness, it presents a large mismatch to the
coaxial feed. Consistent with [451, the active reflection coefficient is normalized using:
76
0.3
•- CAVIARR, E-Planez
0.25 -... CAVIARR, H-Plane0
.. [] PARANA, E-PlaneILOU 0 PARANA, H-Plane
0 0.0
o0
u_
w 0.1
C 10 20 30 40 50 60
SCAN ANGLE (deg)
Figure 45. Rectangular Waveguide Array Active Reflection - Comparison of ResultsUsing Cavity Array (CAVIARR) and General Array (PARANA) Models
.> •.15
S.06
5 .
Figure 46. Microstrip Patch Radiator (dimensions in wavelengths)
77
Ra(,4.) -Zi.(O,•) -zi.(0,0) (5Ra (0, Zi) (=Zl 010 (65)Zi, (0, )+Zi, (0, 0)
This measure discounts the effect of the feed impedance mismatch to more clearly reveal the
effects of scanning. Figure 47 compares the PARANA (HFEM) and moment method calcula-
tions. The agreement is within approximately 7%, with the greatest difference at the angle where
there is a scan blindness due to a surface wave in the dielectric slab. This discrepancy may be
due to the feed modeling ([451 used an idealized probe feed). The fact that the PARANA code
is predicting the existence of the trapped surface wave is a further confirmation that the periodic
boundary conditions are effective and correctly implemented.
9.4. Clad Monopole Array Experiment
A candidate antenna design for a space-based radar, proposed by Fenn, was a planar
array of monopoles above a ground plane (461. Each monopole is simply an extension of a
Zw
0o.8
U-w0o 0.6z0
W 0.4..J
--
- ozar (MoM)
0 - . HFEM
0 10 20 30 40 50 60 70 80 90SCAN ANGLE (deg.)
Figure 47. Active Reflection Coefficient for Microstrip Patch Array, E-Plane Scan
78
coaxial waveguide's center conductor. Fenn's MoM analysis was adequate for the bare mono-
pole, but not for one with a dielectric sheath, or cladding. The sheath, which can be simply an
extension of the coax insulator, is an important enhancement to the monopole design because it
increases the bandwidth substantially.
9.4.1. Initial Validation. For a first validation check, PARANA calculations were
compared to Fenn's calculations for .25Xo-long monopoles arrayed in a .5Xo x .5Xo lattice.
Figure 48 shows the active reflection coefficient vs. scan angle. The essential scanning charac-
teristics are verified, although there is some discrepancy at the angles where the reflection coef-
ficient is very small. This is at least partly due to a simplification in Fenn's MoM model: It
used an assumed form of current distribution (piecewise sinusoidal) on the monopole that is not
allowed to change with scan angle. Herper and Hessel used the same simplification, and showed
a large disparity between calculated and measured results in the vicinity of 600 1471.
9.4.2. Bandwidth Enhancement with Cladding. Figure 49 shows one half of the tetrahe-
dron mesh used by PARANA for the following test case: The monopole (15.875mm long) is
represented as a void extending through the mesh from top to bottom. Shading has been added
to the figure to identify the cells comprising the dielectric cladding (Teflon, Er= 2 .1). The
remaining mesh cells are free space. The lattice is triargular with dx=36mm and dy=31mm
(nearly equilateral). This interelement spacing allows scanning over the 0<900 hemisphere
without grating lobes for frequencies up to 4.8 GHz.
Figure 50 shows contour plots of I R vs. scan aný', and frequency for clad and unclad
monopoles. These calculations are for the (A=90' scan plane, but similar results would be ob-
served for any scan plane due to the equilateral lattice. Both arrays achieve an acceptable reflec-
tion coefficient (IRa I <:.33, or VSWR <2:1) over only a very limited range of scan angles near
79
waD-
Z 0.8
0.8W*
LLL-
00.4 4
z 40
S0.2W- Fenn (MoM)-JLL PARANA
00 10 20 30 40 50 60 70 80 90
E PLANE SCAN ANGLE (deg.)
Figure 48. Reflection Coefficient vs. Scan Angle (0k=0 plane) for a Monopole Array
z
Figure 49. Finite Element Mesh for a Clad Monopole in a Triangular Lattice
80
(I(II[
5.3 r \ J B I
gr
z4.8 ----
------ --- (a )
3.8
35 45 55 65 75 85
SCAN ANGLE (deg.)
S4.8 7 ;, ;' " • --BAND- _V
Z - WIDTH
-\ - . _.-- _ . _::. _
"a 4.3--,---1------ ---._-:- --- ----- ,---;,'- -- i- I- "- - A
35 45 55 65 75 85SCAN ANGLE (deg.)
Figure 50. Active Reflection Coefficient Magnitude vs. Scan Angle (horizontal axis, deg.)and Frequency (vertical axis, GHz) for Bare (top) and Clad (bottom) Monopoles
81
'II' \ I\\\\III
0=60'. The clad monopole array appears to have nearly twice the bandwidth of the unclad array
at 0=600. This simple example case demonstrates the usefulness of this code for investigating
radiator designs - the improved (clad) radiator could not be simulated with existing method of
moments codes.
9.4.3. Experiment. A 121-element clad monopole array was fabricated using the same
lattice parameters (dx=36mm, dy-=31mm) in an isosceles lattice. The monopole element
simply an unmodified SMA connector. When installed in the 1/8"-thick Aluminum backplane,
the conducting monopole and dielectric cladding extend 14.7mm and I 1.8mm, respectively,
above the ground plane.
The array layout is sbown in Figure 51. The measurement procedure consists of connect-
m - -7 -6 -5 -4 -3 -2 -1 0 1 2 3
n i 3. ' " 9" .' ( .. 4 o . 9 " .
. .. . . . . .r .
*0" 9 0 ; 9; " 0 "" 9
REFERENCEELEMENT
Figure 51. Experimental Array Geometry used for Coupling Measurements
82
ing the center element to port #1 of a network analyzer, then connecting each other element in
turn to port #2 and measuring the coupling coefficient C,,,
The active reflection coefficient as a function of angle is given by
Ra(O',4) - E E Cmn e-Jx(inax+ ndyc~toY)e-Jfyndy (66)m=-o n=-oo
Once a finite number of Cm.'s are found by measurement, an approximation to Ra may be
computed for all angles. Figure 52 compares measured results and PARANA calculations. The
oscillation (with angle) of the measured data about the actual Ra was expected due to the finite
array size.
These results, combined with the earlier waveguide and microstrip patch results provided
the validation for the PARANA code. The final two radiator designs discussed next are attempts
to exploit the code to assess the properties of radiators for which results are not available by other
computational methods.
9.5. Printed Dipole Radiator
9.5.1. Element Design. The design for the dipole element shown earlier in Figure 2
follows general guidelines given by Edward & Rees [6]. Its height is .25;o; its overall length
is .5Xo and its arms are .05Xo wide. The design center frequency is 4.8 GHz, giving X0=
62.5mm. The substrate material chosen for this case is 50 mils thick with relative permittivity
of 10 because such material is readily available and it provides a good scaled representation of
1001im GaAs (Gallium Arsenide) at millimeter wave frequencies.
The actual dimensions used for this test case are shown in Figure 53. Figure 54 is an
exploded view of the tetrahedron mesh used as the input to PARANA, showing the three mesh
regions corresponding to the substrate and the two air regions on each side filling the unit cell.
83
l0.8
Z 0W 0.6
LL 0U-o 0.40
0I-- 0.2 Measured0U-J 0 Calculated 0U- [LL.I 0 I I I t I
0 10 20 30 40 s0 60 70 80 90y-z PLANE SCAN ANGLE (deg.)
Figure 52. Measured and Computed Active Reflection Coefficient vs. Anglefor Clad Monopole Array Experiment
The array lattice is square, with .5Xo inter-element spacing. Note that the mesh is denser in the
dielectric slab (by a factor of Er 112 ), and gradually relaxes going out towards the sides of the unit
cell. The dipole and balun are metallized (or photoetched) on opposite sides of the substrate.
The slot in the dipole center divides that part of the structure into two coupled microstrip lines.
They are to be driven 1800 out of phase by the balun. The dipole width is chosen as three times
the balun width so that it provides a ground plane for the microstrip lines comprising the balun.
In the design shown in Figure 53, the first arm of the balun is a linear taper from microstrip line
widths corresponding to 500 (the same as the coaxial input) to 800 (the narrow end). The second
and third balun arms must have the same characteristic impedance as the coupled microstrips, and
values below 800 require extremely narrow slots that are difficult to construct given the toleranc-
es of photolithographic processes. The slot length is approximately 1/4 guide wavelength from
its closed end to where it crosses underneath the microstrip line. Similarly, the microstrip line
84
___________ 31.25mm
W... ....... ....m m
Width=1.19mmSlot WidthzA8mm ....
Figure 53. Printed Dipole Radiator Design
Figure 54. Exploded Finite Element Mesh for Printed Dipole Radiator
85
length is 1/4 guide wavelength from the slot to its open end, but it is reduced by an effective
length using an approximate formula due to Hammerstad [481 (see also [49:1901).
9.5.2. Calculations. Initial calculations of Ra vs. frequency revealed that the design in
Figure 53 was a poor radiator, with IRaI greater than .8 at the design center frequency (4.8
GHz). Slightly better results were obtained by scaling the height by .8 so that the overall height
was 12.5mm (.20XO) and then reducing the dipole width from .4Xo to .33Xo. Figure 55 shows
I R.I vs. frequency for both dipole widths with the 12.5mm height. Unfortunately, this printed
dipole has not been tested as an isolated radiator on high-permittivity substrates, so it is not
known how much of the mismatch is due to array effects vs. feed line mismatches. Nonetheless,
the scanning properties may still be evaluated, normalizing the active reflection coefficient using
(65). The results for 4.8 GHz, shown in Figure 56, indicate that there are no scan blindnesses
in either the E- or H-plane (the E plane is the plane containing the substrate).
1
zwC 0.86ILuJ0oo 0.6Z0o 0.4
LU Length
w 0.2 I> •- 40X.40 Length.2 -•k, .33 X0 Length
0 _ _
0 I I f i . .......- - - - -
4 4.2 4.4 4.6 4.8 5 5.2 5.4
FREQUENCY (GHz)
Figure 55. Computed Active Reflection Coefficient at Broadside Scan for Reduced-Height(12.5mm) Printed Dipole for Two Dipole Lengths
86
1I___I___I____I___I______________
Wa -*- .33 Length, H Plane
t- [] .33 Length, E PlaneZO0.8 __ _ _ _ _ _
0
S0.6 -U.
1U.0 0.4o0
o0 0.2-
-J9X_ 1 - --- 40
0 10 20 30 40 50 60 70 80 90
SCAN ANGLE : (deg.)
Figure 56. Computed Active Reflection Coefficient (Normalized) vs. Scan Anglefor Reduced Height (12.5mm) Printed Dipole, 4.8 GHz
9.6. Flared Notch Radiator.
9.6.1. Element Design. The basic idea behind the flared notch radiator shown earli,ýr
in Figure 1 is a slotline, gradually opening out to provide a tapered impedance match to free
space. There do not appear to be any specific design rules 150], but generally, the longer the
flare, the greater the bandwidth. For purposes of this study, the exponential flare given by
Choung & Chen was selected [511.
Figure 1 showed the slotline being fed by a microstrip line, which is in turn fed by a
coaxial cable. This is a form of balun (the same arrangement used for the dipole in the preceding
section) matching the balanced coax to the unbalanced slotline. An alternative balun design is
based on a new coplanar waveguide (CPW)-to-slotline transition that terminates one side of the
CPW in a broadband open circuit 152]. This design has the advantage that only one side of the
87
substrate is metallized, reducing the number of steps in fabrication and eliminating the possibly
of registration error. The metallization pattern for a single flared notch radiator is the shaded
area in Figure 57. The test case used a 50-mil thick substrate with relative permittivity of 6.0.
The flare length and mouth width are 33.3mm and 30.0mm, respectively. The flare shape is
given by
w (z) = woexp -.z. In [WO (67)
where wo and wm are the widths at the slotline an the mouth and L is the flare length.
9.6.2. Array Performance Calculations. One of the geometry models used as the input
to PARANA is shown in Figure 58. As was the case with the dipole element, the mesh is denser
in the substrate than in the free space regions. The unit cell size is dx=36mm and dy=34mm
(rectangular lattice). A second model (not shown) used an equilateral triangular lattice with
dx=62mm, dy=36mm and,= 60 '. Both of them used feed waveguide dimensions corresponding
Wm
METALLIZATION
OPEN CIRCUIT
YFigure 57. Flared Notch Element Design
88
SCOAXS• APERTURE
SUBSTRATE
GROUNDPLANE FACE SUBSTRATE SURFACE MESH
(a) (b)
Figure 58. Finite Element Mesh for Flared Notch Radiator (Rectangular Lattice):(a) Unit Cell Showing Coax Aperture; (b) Substrate Surface Mesh
to APC-3.5 coax (a=.75mm, b= 1.75mm).
The broadside (0o=0) active reflection coefficient is shown as a function of frequency in
Figure 59. These indicate that the radiator is capable of very broad bandwidth, but the skewed-
lattice array has a resonance effect that causes a blindness near 4.25 GHz. The rectangular-lattice
array is a very promising design, since its predicted bandwidth of greater than 50% is difficult
to achieve in an array.
The active reflection coefficient vs. scan angle for the rectangular-lattice array is shown
in Figure 60. From the low (3GHz) to the center (4GHz) of the frequency range, it behaves
well, but at the high end (5GHz) it displays blindnesses in both scan planes (the E plane is the
plane containing the substrate).
89
1- Rectangular Lattice
Z-0- Triangular Lattice 0
~0.8-L6~
00 0.6, 'z 0O ,60
-- 0.4 '0&L.
> 0 .2 - 0
010
2 2.5 3 3.5 4 4.5 5 5.5 6
FREQUENCY (GHz)
Figure 59. Active Reflection Coefficient vs. Frequency for Printed Flared Notch Arrays
9. 7 Summary
The results of this chapter have proven the validity of the essential feature that makes the
hybrid finite element method applicable to infinite array analysis: the periodic boundary condi-
tion implemented by "wrapping" opposing unit cell edges onto each other with an appropriate
phase shift. It was successful for both rectangular and trapezoidal unit cells, the latter applying
to skewed array lattices. The microstrip patch array test case showed that it correctly predicts
the behavior of surfaces waves in a dielectric layer on a ground plane. The monopole array test
case further demonstrated the method's ability to deal with inhomogeneous dielectrics. Most
importantly, the same computer program with no changes whatsoever executed the computations
for every one of the seven separate array/radiator geometries discussed in this chapter. The only
things that changed were the finite element model created in I-DEASTM and the user instructions
90
1wa -- 3.0 GHz
I-- 4.0 GHzz 0.8 - 5.0 0Hz
I-
zZ (
LU 0.6
(a)LL.w
~0.4-z0
W 0.2LLU-
0 rI I P
0 10 20 30 40 50 60 70 80 90
SCAN ANGLE O (deg.), E-PLANE
w 1Q
E-0 3.0 GHzI-- S--• 4.0 GF~z0 0.8 - 4
<• 5.0 0Hz
,' (b)
z0
IL.uJ
0 I-
0 10 20 30 40 50 60 70 80 90
SCAN ANGLE 9 (deg.), H-PLANE
Figure 60. Active Reflection Coefficient vs. Scan Angle, Printed Flared Notchin Rectangular Lattice: (a) E-Plane Scan; (b) H-Plane Scan
91
identifying the feed waveguide type and location and the array lattice parameters. Thus it is
demonstrated that the finite element method has unparalleled versatility for phased array analysis.
The question of efficiency is addressed below in Table II, which summarizes the time and
storage requirements for most of the test cases discussed in this chapter. The total run time
(tabulated figures are for a VAX® 4400 mincomputer whose performance is rated at about 17
MIPS) is dominated by the matrix solve time, which was as high as 7 hours per point for the
printed dipole (the most time-consuming case). The most time-consuming part of the matrix fill
is the calculation of the radiating aperture terms. It is highest for the microstrip patch because
that case had relatively more edges in the aperture than did the other cases. The fact that these
computations could be performed on a typical minicomputer are encouraging, although whether
or not one regards them as "efficient" depends on the difficulty of solving the design problem by
other means.
Table II. Mesh Size, Execution Time and Matrix Storage for PARANA Test Cases
Test Unit Cell Mesh Mesh Iter- Fill Solve MatrixCase Vol. (X3) Cells Edges ations Time Time Size
(mrin.) (min.) (Mbytes)
Microstrip Patch .01b 4291 4862 2500 44.5 198 6.2
Clad Monopole .063 3092 3487 19000 7.1 201 0.7
Flared Notch with .17 4443 4778 18000 8.7 124 0.8Triangular Lattice
Flared Notch with .19 4845 5271 23500 9.5 196 0.9Rectangular Lattice III_
Printed Dipole .05 6659 7447 39000 20.6 428 1.6
92
X. Conclusions and Recommendations
This research project began with a novel concept for modeling infinite phased arrays and
concluded with a demonstration of its capability. The work in between involved the entire
process of electromagnetic predictive code development: casting the physical problem as a
mathematical boundary value problem; mapping the latter to a linear system using finite element
and moment methods; designing, writing and troubleshooting the general-purpose computer
program; adapting the commercial software for geometry generation; and finally, validating the
code. Many difficulties needed to be overcome, both expected and otherwise; yet, other expected
problems proved inconsequential. This chapter attempts to summarize those findings and to
assess the implications of the results to future work.
10.1. Conclusions
10.1.1. Theory and Formulation. Thv successes of other researchers in applying the
finite element method to time-harmonic electromagnetic problems was reason to believe that it
would also succeed for the phased array problem. Nonetheless there are always doubts su;-round-
ing any such implementation given that the problem does not have the properties of self-adjoint-
ness or positive definiteness. The attendant risk that the iterative matrix solver might converge
to a false solution, or not converge at all, did not materialize. The conclusion is that the weak
form and Galerkin's method are appropriate to this class of problems.
There have not been any cases that would indicate instabiity or non-uniqueness, which
are always concerns when the properties of fields in two or more regions must be met. A unique
solution generally requires that continuity of tangential electric and magnetic fields must be
independently enforced. The fact that this problem involves boundaries that are planar and
exterior solutions in terms of discrete modes whose tangential field components are dependent
93
evidently circumvents that requirement.
10.1.2. Implementation. Three-dimensional finite element problems involve very large
systems of equatio, s. As seen repeatedly in the previous four chapters, even relatively small
devices result in grids with thousands of edges, usually because of the need to capture fine e -tails
in the geometry. The uncertainty of whether the available computers could solve the resulting
matrices in a reasonable time has been resolved and the validation tests indicate that most practi-
cal array radiators may be analyzed using typical minicomputers.
The three codes TWOPORT, CAVIARR and PARANA are geometry-independent, within
the constraints of the generic classes of problems they are designed to solve. They exploit the
commercial CAD software that benefits from decades of research oriented towards mechanical
engineering applications. Two observations regarding that software are: (a) it is capable of
generating grids suitable for phased array analysis; and (b), of much greater significance, it
removes the geometry generation from the electromagnetic problem so that the codes may encom-
pass far broader problem classes than previously attei;.,ted.
10.1.3. Validation. The extensive set of validation cases that were used to test the three
computer codes demonstrate the effectiveness of the key elements of the solution approach. The
interior finite element solution accurately incorporates electromagnetic boundary conditions at
perfect conductors (including sharp edges) and dielectric interfaces. It is evidently free of spuri-
ous, non-physical solutions, indicating that the divergence condition is also satisfied.
The waveguide mode integral equation was shown to be an effective mechanism for
enforcing field continuity at waveguide apertures. Its payoff is in providing an accurate means
for modeling antenna feed structures.
The hybridization of the periodic integral equation with the finite element solution was
94
shown to correctly enforce the radiation condition above the infinite array. Its implementation
ion in the CAVIARR code allowed the correct prediction of active reflection coefficient for a
variety of radiator types, including cases involving scan blindnesses.
The implementation of a periodic boundary condition on a three-dimensional finite
element problem is evidently a first. The success of the final stage of this work hinged entirely
on that single unproven algorithm and the question of whether or not it would conflict with the
periodic integral equation. Thus, the most important finding is that the algorithm involving
boundary "wrapping" works as proposed, and complements the periodic radiation condition.
10.2. Recommendations
The recommendations fall into three broad categories. The first deals with hardware
experiments needed to demonstrate the potential for design, in contrast t(. the analysis that was
the main subject of this project. The initial designs for the printed dipole and flared notch
radiators were a first cut, and their performance leaves much to be desired. Further experiments
will be necessary to: first, perfect the design for single, isolated radiators; second, identify the
geometry parameters that influence their impedance match in the array environment; and finally,
use the codes to optimize those parameters for best performance over some specified range of
scan angles and frequency.
The second category deals with possible improvements and enhancements to the computer
codes. For example, the TWOPORT code could easily be extended to deal with multi-port RF
devices. The inclusion of that feature in PARANA would allow the simulation of radiators that
have more than one feed port, such as dual-polarized and multiple-frequency antennas. All three
codes would benefit from faster matrix solution, which could result from improved iterative
methods, perhaps using preconditioning.
The third and final category of recommendations deals with related or similar electromag-
95
netic problems that may benefit from application of hybrid finite element methods. One that is
a particularly straightforward extension of the present work is gratings and frequency selective
surfaces. That problem may be addressed simply by replacing the waveguide feed with a second
periodic radiation condition and an incident field in the form of a plane wave. A more difficult
extension, but one with iiany practical applications, is a finite-by-infinite array. Such a model
could be used to predict the radiating properties of line-source arrays or to assess edge effects in
planar arrays. Last, the calculation of scattering from objects that include cavities, or of coupling
into circuitry inside those cavities are problems that may benefit from the use of the finite element
method to model the cavity interior, with an integral equation boundary condition at the aperture.
These problems are more easily addressed now because the present work has provided, among
other things, a body of well-tested computer routines for three-dimensional finite element and
waveguide mode computations.
96
Appendix A
The Electric Field Functional
A. 1. Variational Principle vs. Weak Form
Publications dealing with electromagnetic finite element applications usually begin with
one of two forms of functionals2 with little or no discussion as to why one is used and not the
other. This appendix discusses their origins and the circumstances in which each is appropriate.
The conclusion is that the results of the two methods are indistinguishable for typical electromag-
netic radiation and scattering problems.
The first alternative is
F, (E) = 2J VXJJVxEk orE E dv +jkoqoJ(EXH).tAds (A.1)
al F
(see, for example, Jin & Volakis [531) where Q is the volume region over which the unknown
field is to be found and r is its enclosing boundary. The subscript v denotes that this functional
is the variational principle.3 The second alternative is
Fw(E) = 1J [VX .Vx _.koEr W.E dv +jkoiio (WxH) -Ads (A,2)Ar 11
(see, for example, D'Angelo & Mayergoyz 1241). Here, W is a trial function whose form is yet
to be determined. The subscript w on F denotes the fact that this is the weak form.
2 A functional is a mapping from a space of functions to the complex numbers. It is usually
an integral containing an unknown function. The result of the integration is a single number, incontrast to an integral equation, which maps the function into another function.
3 A variational statement may be either a variational principle or a weak form.
97
The variational principle Fv has been deduced from a general from of energy functional
[541 and is used in conjunction with the Rayleigh-Ritz principle in order to form a system of
equations. The weak form is derived more directly by simply taking the inner product of the
operator equation with the trial function. It will be used in conjunction with the method of
weighted residuals to form the system of equations. In the case of Galerkin's method (a special-
ization of weighted residuals) the two forms may give exactly the same system of equations. But
in order to use Galerkin's method, the expansion function that is admissible in the original
problem must also be admissible in the adjointproblem.
The following section will discuss the meaning of the adjoint problem and what the
conditions are for self-adjointness as applied to the vector wave equation. The third will then
show that typical waveguide continuity conditions represent non-self-adjoint boundary conditions.
The last section will show that under the assumption that Galerkin's method is applicable, the two
formulations generate identical systems of equations.
A. 2. The Adjoint Problem
Before deciding whether to use (A. 1) or (A.2) one must know whether or not the operator
equation is self-adjoint. If not, then the properties of the adjoint problem must be determined in
order to ensure that the trial functions are capable of representing its solutions.
The operator equation is the time-harmonic, source-free vector wave equation for the
electric field in a linear, isotropic, inhomogeneous region:
-I - 2L(E) = Vx-VxE-koE rE= 0 (A.3)Mr
Its inner product with an arbitrary complex function W is
98
<L(E),W> = [VXiVXE-koWr * dv = 0 (A.4)/Xr
Using a Green's identity (integration by parts) twice shifts both derivatives from E to W:
<L(E),W> VJJ [IVXE V xW* -korE' W dvi J (A.5)
- r w*xVxE.i ds
r
E JJ ~.[x Vxiw'k~E:ri']*dv
(A.6)-- VE J-E-xv~X'vx WJ*].h ds
From the definition
(L(E),W> = <(E,La( W)> (A.7)
it is evident that the term in brackets in the first integral of (A.6) is La, the adjoint operator. It
is simply the wave equation with the constitutive parameters replaced by their complex conju-
gates. It is now clear that W must be in the domain of the adjoint operator.
The definition of self-adjointness is L=La, which obviously cannot be true if the problem
includes lossy materials. But it also depends on whether the boundary conditions are such as to
make the surface integral in (A.6) vanish, i.e.
~iIa*xvxE. fds -- xv xE" *.•ds (A.8)r r
EJ ~ a* XH<i.Ads E HJ f~EIIZ ilds (A.9)rA~rr Xr 9 r
99
Suppose that all lossy magnetic materials are confined inside Q so that along the boundary r, the
permeability is entirely real. Then (A.9) is satisfied when Ea=E* and Ha=H*. Recall that the
time-harmonic and time-dependent fields are related by 155:15]
E(x,y,z,t) = / Re(Eejio1) (A. 10)
(boldface represents the time-dependent quantity; the expression for magnetic field is identical).
Evaluating (A. 10) at any point f along r, with Eo denoting E(?):
E(r,t) = v/2[Re(Eo)cos wt- Im(EO)sinwt] (A.11)
and assuming that Ea=E*
Ea(r,t) = F2-Re[EkO eJwt} = F2-[Re(Eo)cosowt + Im(Eo)sinw t]
= v'2Re[Eoe -i•] (A.12)
= E(r,-t)
In other words, the adjoint fields are time-reversed versions of the original fields. They carry
power across the boundary in the opposite direction and they encounter materials that have gain
instead of loss. This is the physical interpretation of the adjoint problem, and is consistent with
the property that if the operator L is causal, then the operator La is anti-causal [56:3561.
Notice that if the boundary r is comprised entirely of perfect electric and perfect magnet-
ic conductors, then (A.9) is always satisfied because there cannot be any transfer of power across
such boundaries. This leads to the suspicion that the boundary conditions that will cause non-
self-adjointness are those open boundaries where conditions of field continuity are to be enforced.
Section A.4 will demonstrate that this is indeed the case for the waveguide/cavity apertures that
are considered in the main body of this report.
100
A. 3. Continuity Conditions for Waveguide Apertures
For the waveguide continuity problem, the boundary r reduces to the waveguide aperture
rA. Taking Ea=E* and HIa=,H* the following form is equivalent to (A.9):
JE* (flxH)ds JE.-(A xH* )ds (A. 13)
PA rA
(again assuming that ,r is real along r). Consider a waveguide whose axis is the z axis and
which joins the volume (t through an aperture in its end wall at z=O. It is assumed to be match-
terminated at z < < 0. The aperture field due to the dominant mode and its conjugate are
E= g°(1 +Co) (A.14)
H = ×xg•Yo(1-Co)(A. 15)H * -- 2 x 0Yo( I - co*
where go is a transverse mode function, Yo is the modal admittance and Co is an unknown
coefficient. The outward normal to 0 is fh=-2, so the left and right sides of (A. 13) give
JJE* *(fixH)ds = Yo(I+Co0 )(1-Co)JJ go0 J2 ds (A.16)
rA "A
E.xffH*)ds = Yo( +Co)( -C)ff 0I12ds (A.17)rA rA
These two are only aqual if CO is real, which is not generally the case. Therefore, the continuity
condition across a waveguide aperture will render the boundary value problem non self-adjoint;
the functional (A. 1) is not appropriate even when there are no lossy materials; and the weak form
101
(A.2) should be chosen.
A. 4. Galerkin 's Method vs. Rayleigh-Ritz
In the main body of this report, the finite element method is used to produce a system
of equations from the weak form functional. The field is represented by a summation of un-
known complex coefficients with known, linear vector functions:
N
E = e. ((x,y,z) (A. 18)S=1
For the expansion functions to be admissible, they must be in the domain of the functional.
Linear functions meet that requirement since their first derivatives are continuous (integrable).
The residual error is defined as R=L(E)-L(E). N weighting functions will be chosen,
each of which is orthogonal to the residual so that (R,*r)=O, giving the system of N equations
0 =E e, V X•Wr"VXs-ko rWrs dvs=1 Ar V-r (A . 19)
r
The choice *,r=ý satisfies the orthogonality requirement, but it is also required that Or be an
admissible expansion function for the adjoint electric field. Starting with the adjoint operator
equation, forming (E,La(W)) and using the Green's identity once gives
=Lu [ x "Vx -ko V E r W E dv
ii (A.20)- J EXVX W* fidsr
which makes it evident that expansion functions for W must have continuous first derivatives.
Therefore, the same expansion functions are admissible in both the original and the adjoint
102
problems. It is this fact that allows Galerkin's method to be employed.
Consider the variational functional (A. 1) with the expansion (A. 18) for the electric field:
N N12v 2' erj e es V r VX s -rr' r* dv
r1 s= r
N (A.21)
+Jk°n°EerJJ ('rXH).flds' r=1,2,...,Nr=1 r
The Rayleigh-Ritz method equates the stationary point of F, to the minimization of the above
with respect to each of the coefficients er, i.e.
aFV(E)bFv(E) = 0 * _-___ = 0, Vr (A.22)a er
Carrying out the partial derivatives in (A.21) gives
= eJVXs 2
S=1 I(A.23)
+iJkoOjj (ý,XH)×fids, r=1,2,...,N
r
which is identical to (A. 19) with *r replaced by ýr" This shows that the systems of equations
resulting from the variational principle and from the weak form are identical. Thus, under at
least some circumstances the distinction between the two forms is inconsequential, and the
variational principle may also be used.
103
Appendix B
Waveguide Mode Function Inner Products
B.1. Approach
The waveguide interaction terms require the computation of two surface integrals ý%i and
*'. from (36) and (37) where s and i are the edge and mode indices. Each may be found by
summing the contributions from the individual triangles that share edge s, e.g. for triangle k:
•(k) =L,4ksi = '-k ) f)[gix(f2TT3 -fAT2 3)+giy(AT 22 -f2 T12)dxdy] (B3.1)
Ak
Ls is the edge length, ft is the outward surface normal (into the waveguide), Ak is the triangle
area, f, and f2 are the linear scalar finite elements associated with the nodes bounding the edge
and T is the 3x3 simplex transformation matrix for the triangle. When the integral is trans-
formed, dx dy -- 2Akdtldt2 and
Si L5 (a "t)[-T23 G. +1T13 G-2 + T 22Gl -Ty 2 G12 ] (13.2)
1 1-t1G'j= jat, I gi, tj dt2' , =x,y j=l1,2 (B.3)
0 0
Similarly,
s Ls[ T2 2 G. 1 T12 G. 2 + T 2 3 G 13 Gy2] (3.4)
The generic procedure used to compute the terms of the matrix SJE is outlined in Figure Bl. The
bulk of the computation is in step 2.b.ii, where the integrals Gj are calculated. In the case of
rectangular waveguide, they may be evaluated in closed form. For the other two waveguide
104
FOR EACH APERTURE (A,B):
FOR EACH MODE (i):
1. COMPUTE PROPAGATION CONSTANT, MODAL ADMITTANCE
2. FOR EACH TRIANGLE (k) IN APERTURE:
a. COMPUTE SIMPLEX TRANSFORMATION
b. FOR EACH EDGE (s) BORDERING THE TRIANGLE:
i. DETERMINE EDGE VECTOR ORIENTATION
ii. COMPUTE G 6 FOR ,=x,y andj=1,2
iii. ADD CONTRIBUTIONS 4l,i(k) and si,(k) to 4ý,s and 'P/s
3. FOR EACH EDGE (s) IN APERTURE:
a. FOR EACH EDGE (t) IN APERTURE:
i. ADD j ko no Yi 4ý.s *si I to S sE
Figure B1. Procedure for Matrix Fill Calculations Involving Waveguide Modes
types, circular and circular coaxial, they are computed numerically using Gaussian quadrature.
The remainder of this appendix discusses the details of those integrations. Expressions for the
mode functions, cutoff wavenumbers and modal admittances may be found in [171 and [28].
B.2. Rectangular Waveguide
The mode functions for rectangular waveguide will have indices m,n and p, where p= 1
or 2 for TE or TM, respectively. The vector components are
S= Cmnpx Cos ( Osin(n I) (3.5)
mnp = Cmp sin(W) cost !I) (B.6)
where a and b are the waveguide dimensions along the x and y axes, respectively. The normal-
105
iztion coefficients ensure that the modes are orthonormal over the waveguide cross section.
The inverse of the simplex transform matrix, T", will give x and y in terms of tj, t2 and
constant coefficieiis:
x = a0 +aoIti+a 2 t2 (B.7)
Y = 00 + f 1 tl + 02 t2 (B.8)
Let ki and qi represent the following combinations of ai and fi
= yr °tir r[ 39)
m7rea nirfl,+ (B.9O)
•i a b
m Wrai n wr8i (B. 10)
a b
In terms of these, the mode functions may be rewritten:
gmnpx = 'sin(4o+t 1 t 1 ++ 2 t2)-_sin(q0o+qlIti + 2 t2 ) (B.11)Cmnpx 2 2
gmnpy = sin(O+kltl+t2t2)÷2sin(1O"I l + n2/2) (B.12)Cmnpy 2 i(o4t-it)Pnloii7 2 2
Let Hý denote the integral
H I tj dt, J sin(ýO+4Itl + it-2)dt1 (B.13)
0 0
Hl is identical with 71i replacing ýj everywhere. Ht' and H.' are the same, but with I and 42
or "1, and n2 reversed. In terms of these, the integrals in (B.3) are
GxI = Cm.px(H1 - H0) (B. 14)
106
GyI = Cmnpy(Ht +H ) (B.15)
Gx2= Cmnp, x(H - H;) (B3.16)
=Y Cmnpy(H I +H) (B3.17)
Finally Ht may be evaluated in closed form, but it must be accomplished separetely for several
special cases, as given below:
(a) general case:
Hý 1 - 2 12[cs(4°+41 )- cso°+*4 sin(o°+41 )]1 3.18)
1
- i [COS(U 0 +4) -COS(4O+ 2 )+(Ul -2)sin(°+4i](Q I k2)2
s cti2b0, to4 0, a 2r4g IoIn 1 (for small values of t24w) (B. 18) is numerically unstable,
susceptible to large roundoff error and overflow):
Ht {I 2[cos(to+44)_ cos~o +ý4 sin(4o+4,)] (13.19)SI I
-Cos Q 0 + 2 ) +(tI - t 2 )sin([o + Z)I
(c) I421"41:
H4 41= (4k1 +6t2 -tI4 2 )COS(4 0 ) - (44 +642 )C0S(4 0 ÷ ,)44, (B3.20)
-(2k2 +44,4 2 )sin(o) - 2( I+ Z1-,)sin( 0+t,)}
107
(d) I t 1:
Hý coso- It sinto4 (221)
21
-(t1-1 2)A[cOs(4o+4 1)-cOs(4o+4 2 )+(4t-4 2 )sin(4o+4 1)}]
(e) I2 -42)1, 1 l 1 :
Ht = sin 4o + (ti +t)os(o) (.22)
Note that the last four are correct in the limits as those quantities specified as much less than
unity go to zero. Precedence is given to the five formulas in reverse order, checking for condi-
tion (e) first, and executing (a) only when none of the others (d), (c) or (b) are true.
B. 3. Gaussian Quadrature Integration
For circular and coa-i~d mode functions the inner products (B.3) may not be accom-
plished in closed form, so they have been evaluated numerically using Gaussian quadrature.
Quadrature formulas only apply strictly to one dimension, but are easily extended to two dimen-
sional integration over rectangular areas. To use these, the triangle's geometry is transformed
to a unit square, using an approach suggested by Stroud & Secrest [57]. The transformation to
simplex coordinates mapped an arbitrary triangle into one with vertex coordinates are (0. rn). (0,1)
and (1,0) in (t1,t2) coordinates. The second transformation is given by
t2
U, = t1 ; u2 = (1 , ,t 1 (B.23)
I , t1 = 1
The Jacobian of this transformation is (l-t,-), so that a typical integral term transforms into:
108
G,= f 1 dt, Jg(Ijt 2 )dt I = udu I J -u,) 2 g(u 1 ,u2 )du2 (1.24)
0 0 0 0
Let Uk and u, denote the one-dimensional quadrature sample points along ul and u2, respectively,
with Wk and wm the corresponding weights. Then the integral is approximated by the sum
Q QG, E Wk Uk E Wm(l -U.)2 g(Uk,Um) (B.25)
k=1 m=1
where Q is the order of the quadrature formula. See, for example, [571 or [58:8871 for tables
of weights and quadrature points.
109
Appendix C
The Periodic Integral Equation
Conventional derivations for the scanning properties of phased array antennas are often
given in terms of "Floquet modes," which are essentially plane waves propagating in several
discrete directions away from or towards the array. That derivation is an analogue to ordinary
waveguide modes, and was developed as convenient means for deriving mode-matching solutions
to radiation from waveguide arrays. This appendix presents an alternative derivation using
Fourier transforms that may, in some cases, lead to greater insight than the mode-matching
solution.
This derivation proceeds from the integral equation for an arbitrary sheet current. It is
then specialized to the case of an infinite periodic sheet current through the periodicity condition.
Its spatial frequency domain representation is then obtained through straightforward application
of Fourier transform theorems and a Fourier-Bessel transform for the free space Green's function.
The inverse transform then yields the desired result, which is an expression for the integral
equation not as a continuous integral, but as a summation over sample points in spatial frequency.
An extension of this derivation for skewed array lattices is also provided. The results of Galer-
kin's method are specialized to the linear vector finite element functions and analytic expressions
for the resulting inner products are given. Finally, the derivation in terms of Floquet modes is
shown to provide an identical system of equations.
C.). The MFIE for Planar Current Sources
The objective of this section is to obtain an integral equation for the fields in the half
space above the radiation boundary due to the fields below it. The use of the equivalence princi-
ple will simplify the derivation. The boundary, which will be taken to be located at z=O,
110
supports equivalent currents M and J. The source of these are the tangential fields just below
the boundary:
M= x E(z=) (C.1)
J = XH(z=O0) (C.2)
The equivalent problem in the z > 0 half space also sees a conducting boundary, but it supports
-M and -J. This equivalent current is the source of an electric vector potential, F:
-00
:-r) " L J -(-J ) G(-- )dx'dy' (C.4)-00
G(:_W) _ e-kIT'-r (C.5)
I -- J
where r' and r denote, respectively, source and observer coordinates, and G is the time-harmonic
free space Green's function. Here, r' is confined to z=0, but r may be anywhere above z=0.
The magnetic field at the observation point is [59:36]:
HI(G) = -j + VV , F (C.6)jWUE
An integral equation is obtained by applying a boundary condition to the above radiation integral.
Specifically, the total tangential magnetic field at z=0 is H = ×xJ:
- = 21 [ k2F +• + + [ F -a 2 Fx (C7)jxl aX 2 Oxay ay 2 axay
Ill
Note that an electric current source in the z=O plane produces a magnetic field that is entirely
normal to that plane. Hence A(z=O) = 2Az , so J does not contribute to Ht(z=O) and (7) is
a complete expression for the integral equation. The next section will specialize this to the case
in which M is an infinite periodic source.
C. 2. The Periodic Magnetic Field Integral Equation
According to Floquet's theorem, the fields and currents anywhere on and above an
infinite periodic array must obey the relationship
't(x+mdax,y+ndY) = $,(xy)e ei-j~yndy (C.8)
where -t may be any of E,H,M, or F, d, and dy are the lattic spacings in x and y and Vx and Oy
are the phase shifts necessary to produce a plane wave propagating in the 0o,00 direction:
OX = k sin 0o cos4o (C.9)
4'y = k sin 0o sinko (C.10)
Let M• (X,y) denote a unit cell magnetic current that is equal to the source distribution within the
rectangular area -d,•x<x d_ and -dy <y:! dy and zero elsewhere.
Consider an infinite two-dimensional sequence of Dirac delta functions located at lattice
points x =rmd., y=ndy for -o <m,n< oo. Figure Cl illustrates that the effect of a two-dimen-
sional convolution of this Dirac sequency with the unit cell current distribution is to replicate the
current distribution around each of the lattice points. If each impulse is also weighted by the
complex exponential representing the beam steering phase, then M is:
M(x,y) = M, * , ( (x-md,,y-nd.)e -J x e -JYy (C.11)m n
where * denotes the convolution operation. This is an alternative form of Floquet's theorem.
112
UNIT CELLFIELD 2D IMPULSE
SEQUENCE
o
OTAL FIELD
Figure Cl. Two Dimensional Convolution of Unit Cell Field with Dirac Impulse Sequence
(All summations may be assumed to have limits of -oo to + oo unless otherwise noted.)
Let _M denote the two-dimensional Fourier transform of the magnetic field with respect
to the spatial frequency coordinates k, and ky (underbar will be used to denote transformed
quantities):
M(kx'ky) = J J M(xy)e kxxe kyYdxdy (C. 12)-00a
00
= 1rr-jk .,x -jk C.3l(X' y) 2 l [[l(kx,,kyl e- e- Y'ydk,:dky (C. 13)
-OD
The following four properties of Fourier transforms are required (see, for example [30:199-200]):
-9_F * G) (C.14)
113
.riF(x)e-j"x} = F_(k . c) (C.15)
, ,- } = -jkjF (C.16)
21r (k_2*
2 5(x-nd) 1} = (C. 17)
(The last one is a form of the Poisson sum formula.) Substituting (C.9) into (C. 10) and using
properties (C.14), (C.15) and (C.17) results in
Ml(kxk,) = M -•ayL - E 6(kx -kx'ky-kyn) (C. 18)uc d y m nt
kxm - 2 -rm - x (C. 19)dx
kyn ((C.20)dy
The points k.m and kyn are sample points in the spatial frequency, or spectral domain. They may
also be recognized as the so-called Floquet harmonics.
When the source and observer art. both in the z=0 plane the Green's function is only a
function of x and y and the integral (C.3) is a convolution integral written as
F(x,y) = -L M(x,y) * G(x,y) (C.21)4w
whose Fourier transform is
F_(kx,ky) = - -_. M G(k,,ky) (C.22)
The transform of the Green's function is found using a Fourier-Bessel transform [60:121, with
the remarkably simple result:
114
G(kky k2w (C.23)
Using this with the derivative property (C. 16), the MFIE in the spectral domain is
fi xJ1 11 e[ (k2- k.k2,) F -k.,ky~~ F [(k 2-k 2)F -kxkyF 1 (C.24)
Taking the cross product of 2 with (C.24), substituting (C.22) for F, and using (C. 1):
2 k- k; k(k- 1 - (C.25)2kKI) k2k ( 2~k)
= -k2k2-k_2 (C.26)
or in terms of the unit cell electric field:
j = ETE 6 (k,- kx,, ky - (C.27)m n
I (k-k) kxky (C.28)2kT"Lkxky (k2-k)
Finally, testing is to be carried out in the spatial domain, requiring the inverse transform of
(C.25). With the delta function in the integrand, that integration reduces to a sampling of the
integrand at each kxr and kyn:
= Tm nE c(kxm, kyn)e -jk.,x-e .jkV (C.29)m n
This form of the MFIE is only valid for rectangular array lattices, but many actual phased arrays
115
1 k.. , 1(mn = -k 2kinxkynJ
use triangular lattices. The following section will show how (C.29) is modified to accomodate
skewed lattices.
C3. Skewed Array Lattices
Phased array antenna elements are usually arranged in a triangular lattice, formed by
shifting successive rows to the right or left by one half the column spacing. This allows a larger
inter-element spacing (hence fewer elements for a given aperture area) to cover a given scan
region without grating lobes.
Figure 12 shows an even more general case in which the shift between successive rows
is not necessarily one half the column spacing d.. The Floquet condition for this situation is
E(x+mdx+ndycot-y,y+ndy) = E(x,y)e J-jx('nd.,+ndycoty) e-J'y ndy (C.31)
or in convolution form
E(x,y) = E.u(x,y) b '(x-mdx-ndycot-,y-ndY)e -j"xe -j'yy (C.32)mn n
The Fourier transform of (C.30) is required, but it may be obtained without directly performing
the integration. As in the case of the rectangular lattice, a result similar to (C. 16) is expected,
i.e. the unit cell transform times a series of spectral domain delta functions. The Fourier trans-
form pair (C.9) and (C. 16) have a unique interpretation in terms of direct and reciprocal lattices
116
[61:94-981. The coordinates (kx,ky)=(2irm/dx,27rn/dy) are points in the reciprocal lattice corre-
sponding to (x,y)= (mdx,ndy) and (4ir2/dxdy) is the area of one of its unit cells.
In the skewed lattice, the coordinates of any element are integer multiples of the basis
vectors a and b:
b = dc d(C.33)-b I •d.,COt7 + Ydy
The basis vectors in the reciprocal lattice, 1 and # are found by solving
a C1 b . 1(C.34)
a = b oa 0
with the result
(C.35)
(C.36)
The spatial frequency coordinates corresponding to points in the reciprocal lattice are
S2irm (.7kxmn = 27rf - (ma +nf) = (C.37)x dx
Iky/mn = 2 -9 (m f + ni)= 2rn _ 27rmcoty (C.38)kym : Ty mt )-dy_ d,
The primes signify that these are the unscanned lattice points. The unit cell area in a skewed
lattice is the same as in a non-skewed lattice with the same d. and dy, so the spectral domain unit
cell area is also the same. The end result is that the Fourier transform of (C.32) is
117
f k y = E c- -T E 6(.--tnk-yn C.39)X Ym n
k, m n = 27m - k0sin 00 cos o0 (C.40)
kymn 2wrn _ 2irmcot-f - kosinO0osin•° (C.41)
The effect of beam steering is to shift all of the lattice points. The last three equations above
explain the origin of the Floquet harmonics for a skewed lattice, and agrees with the result given
by Mittra et. al. [62:15961. The relationship between 2 xH and E is still as given by (C.29),
except that the sample points are now k.,kym instead of kxm,kyn.
C.4. Expansion Function Fourier Transform
The electric field within the unit cell is expanded in known vector functions ý. with
unknown complex scalar coefficients e.. By linearity of Fourier transforms, the unit cell field
is
N
9u,= ess(kx,ky) (C.42)
where s is the two dimensional Fourier transform of ý.- These may be evaluated analytically
with the help of homogeneous coordinates within the triangles subdividing the radiation boundary
(the faces of those tetrahedra bordering on the boundary).
Suppose that within triangle k edge s goes between local nodes i and j. Then in terms
of the 2D homogeneous coordinates,
. L t (C.43)
118
Let ri denote the Fourier transform of the scalar linear finite element defined at node i, denoted
fi. In the homogeneous coordinates, fi=ti and
00 1 1 -2Ti = I t eti eJkyy dxdy = 2A I dt 2 I ti ejkýxeJkyY dt, (C.44)
-00 0 0
The factor 2A is the inverse of the Jacobian of the transform from (x,y) to (t1 ,t 2 . The inverse
coordinate transform is expressed in terms of six constants given in Appendix B, (B.7),(B.8).
Let B1 denote the following combinations:
B, = ,l kx + 01 ky 1 = 1,2,3 (C.45)
Then substituting into (C.44):
o jB -t2
Ti = 2Ae A eJB2t2dt2 j tie JBt"dt1 (C.46)
0 0
Five separate cases must be considered, depending on the range of the constants B1.
(a) general case
=2AeJBo -__j + jeJB2 + e JB(jB2 -2jBI +BIB 2 -B2) (C.47)
B B2 B2(BI -B2)' , B - B2)2
(b) 1131-132 < <I1
2AejBO J 1e jB2 A + B B2)2 - [2 2 3 12B2 (C.48)
+ B l-eJ'(I -jB 1 ) }
119
(c) B2 1«< <1
2AejB2 {I -ej" [( 2 B, + 3B 2 ) + B, (BI -B2) ]IB 4 (C.49)
S[j(2B, + 3B 2 - 'B B2 ) -B,(B÷+2B2))
(d) B111<<<
2AeJBo fEB(B 2
- 4 (E j2 (JB!+2 )+B 2 (C.50)
(2BB 2 +,6B2 -. 'BIB 2 ) ÷j(-2B, -B 2 +÷BB2B + 'B,)
(e) IBI1<<l and I1B1<<1
=2 AeJBoTI = 120 (20 + IOjB1 +5jB2 -2BB 2 ) (C.51)
The small argument forms ensure numerical stability. The expressions for -r, are obtained by
reversing B1 and B2 .
C.5. Integral Equation from Floquet Modes
The following is a more conventional derivation of the periodic integral equation using
"Floquet modes." It follows the same general procedure used in deriving the integral equations
for waveguides.
The orthonormal vector Floquet modes 129:41-42] are analogous to waveguide mode
functions:
- ln kyn-•kxm] ei(k'•px+ky) (TE) (C.52)
120
2mn 1 _kxra + _k_ eJ(kxmX+kyy) (TM) (C.53)
d xdy -_k_ xm + ky2 In
The corresponding modal admittances are
KpnKmnlkoiJo p= I (TE) (C.54)Ypmn / kocmn 1)0 p=2 (TM)
Using the waveguide integral equation from Chapter IV (29) but expanding the mode sum over
three indices:
oaoa2
E Ypmngpmn I E•t gpmn ds - J = 0 (C.55)m=0 n=1 p=1 IrR
Et is the transverse (to z) unit cell electric field on rR. Due to the form of the complex exponen-
tial factors in (C.52) and (C.53), the integral (C.55) results in factors involving its Fourier
transform, E. The TE and TM mode terms from (C.55) are
Yimn mnJ Et-gimn ds - Kmn/k0onoIRt dxdy(k-m+k-n)
(C.56)
[fE k~ 2 I~k yn -y k 2E,~
Y[m ,d g2m , E *• " 2mn ds k K.nO °
dxdy(kxm+k)
[Exk Xm +_ykxmkyn + Ex n + Yn]
Summing the TE and TM modes:
121
Y2pmn gpmn J E ,. as = El *._____Kmnilokodxdy (C.58)
•EX (ko -](,yn) + XE y kxmkyn + YE x kxmkyn + EY9Y(k(O - m]
When this is written in dyadic notation, it is clear that the integral equations (3.29) and (B.55)
are the same.
122
Appendix D
Periodic Boundary Conditionsfor the Finite Element Problem
This appendix develops the method for applying periodic boundary conditions to a finite
element problem in one dimension. Consider the two functions f(x) and h(x) shown in Figure
Ih(x)I
Figure D1. Periodic Functions
DI, related by a linear operator equation Lf=h. Their magnitudes are periodic, repeating on
each interval , but each interval has a progressive phase shift relative to the next:
f(x +nd) = f(x) ej n (D. 1)
This may be regarded as a periodic boundary condition for the function on the interval I0,d].
For example purposes the method of weighted residuals will be used to produce a functional:
F(f) = <Lfw> - <h,w>
d (D.2)I [Lc()w h hw*]dx
0
The functions f, h and w will be represented as sums of complex coefficients (fi , hi , and wi)
times scalar basis functions ti(x) where i may range from - o to + co. N + I of these expansion
123
t 1 (x) tN(X)
A i AA A A
i/ S \\ S
-1 0 1 3 N-2 N-1 N +NI N*2(x-0) NODE INDEX (x-d)
Figure D2. Expansion/Weighting Functions
functions are nonzero within the interval 10,d] as illustrated in Figure D2. f, and h, are the
values f(0) and h(0); and fN and hN are the values f(d) and h(d). The functions ti(x) are not
necessarily linear as shown, and do not necessarily extend over subintervals of the same length.
However, those in successive intervals must be replicas of each other with ti+N(x)=ti+I(x-d).
After expanding the functions in (D.2), the derivative of F with respect to each wj gives an
infinite tridiagonal system with matrix and right hand side entries
0o
sj' i =I LlItj (x) Itj (x) dx (D.3)
gJ I h (x) tj (x) dx (DA).4
-00a
For example, the equations pertaining to the nodes -l1,0, 1,2,..., ,N -! 3 are
124
S. . / I / I I/
so,-I f- I + So'o A + so, IfA = go
Sofo A + + s1,2A2 = 1
s2 ,1 fA + s2 ,2 f 2 + s2 ,3 f3 = 92
(D.5)
SNI-,V-2N2-2 ÷ SN- INA-fN-I + SN- ,NfN = gN-I
SN,N.-I fN-I + SN, NfN + SN,N+I fN+I = gN
SN+I,NfN + SN÷INlJfN+I + SN+IN+2fN÷2 = gN+I
The periodicity conditions on the discrete coefficients f and g are
fi+N-l =fiej4' ; fj-N+l =fie-jý (D.6)gi÷N-I = giejV; ; gi-N+i = gie-j(.
The matrix elements must also satisfy a periodicity condition. Since %+1 and ti+N are identical
for all i, it is clear that sj+I , j+1 = si+N, j+N. We can rewrite the system so that it only
involves the unknown values of f and known values of g within the interval [O,d):
SNI,N2fN_2e-J" + sNl',NV-lfN-le-j' + NN-I.NfI glV-e-jý (a)
sNN-NfN-je-JA + S1,1fA + S1,2 = g1 (b)
S2, 1fA + S2 ,2 A2 + S23 f3 = g2 (C)(D.7)
SN-I,N-2fN-2 + SN-1,N-I fAN-I + SN-I,Nfte" = g9N-I (d)
SN,NIfN-I + s1,,1feye/ + S1 ,2 f 2 eJ'b gle)j' (e)
s2,1 fl ejI + S2 ,2 .f2el . + s 2 ,3 f 3 ej# = g2 ej•" ('f
Multiplying (b) and (c) by i0 and subtracting from (e) and (t). repsectively, eliminates the latter
three from the system. Similarly, multiplying (d) by ejA and subtracting from (a) eliminates (a).
125
Continuing the process will eliminate all equations preceding (b) and succeeding (d) leaving only
a reduced system of N equations:
SN,N-IfN-Ie-J" + S1lf 1 + Sl, 2 f 2 = g1
S2 ,1 f + S2, 2 f 2 + S2 ,3 f 3 = 92 (D.8)
SN-I,N-2fN-2 + SN-1.N1fN-1 + SN-I,Nflej"t = gN-1
Suppose that now the original problem geometry is truncated at the boundaries x=0 and
x=d, and the two ends are "wrapped" back on each other, as shown in Figure D3. Now nodes
t 1 (X) t N (X)
>1 x•\ // X
N-2 N-1 1 2 N-2 N-1 N 2 3
(x-O) NODE INDEX (x-d)
Figure D3. "Wrapped" Domain
1 and N are the same point, as are 2 and N+1, etc. Th n the inner product in (D.3) has new
terms for (ij)=(1,N-1),(N-1,l), and referring to (D.8) it is evident that
SI,N-I = SN,N-Ie-j (D9)
SNI ,1 SNI,Nejý
This system is no longer tridiagonal because the boundary terms have introduced new elements:
126
$I, S1,2 0 0 . . . 0 0 SN,NI e
S2,1 S2 ,2 S2,3 0 * * * 0 0 0
0 S3,2 S 3 ,3 S3 ,4 * 0 0 0
tsI=(D. 10)
0 0 0 0 ' • * SN-2,N-2 SN-2,N-1 0
LSNI,NeiO 0 0 0 . . 5"N-I,N-2 SN-1,N-1 SN-I,N
If the original (infinite) matrix was Hermitian (adjoint), then SN,N1 = S*N1,N, and the new
system is adjoint as well. This indicates that periodic boundary conditions do not necessarily
cause an operator to become non-self-adjoint.
127
References
1. Stark, L., "Microwave Theory of Phased-Array Antennas - A Review," Proceedings of theIEEE, 62, pp. 1661-1701, Dec. 1974.
2. Mailloux, R. J., "Phased Array Theory and Technology," Proceedings of the IEEE, 70, pp.246-291, Mar. 1982.
3. Schell, A.C., "Trends in Phased Array Development," Phased Arrays 1985 SymposiumProceedings, Vol. 1, Rome Air Development Center: Hanscom AFB, MA, RADC-TR-85-171,pp. 1-6, Sep. 1985.
4. Lewis, L.R., M. Fasset and J. Hunt, "A Broadband Stripline Array Element," Proc. IEEEAntennas & Propagation Int'l Symp., Atlanta GA, pp. 335-337, Jun. 1974.
5. Cooley, M.E., D.H. Schaubert, N.E. Buris and E.A. Urbanik, "Radiation and ScatteringAnalysis of Infinite Arrays of Endfire Slot Antennas with a Ground Plane," IEEE Trans. Anten-nas Propagat., AP-39, pp. 1615- 1625, Nov. 1991.
6. Edward, B. and D. Rees, "A Broadband Printed Dipole with Integrated Balun," MicrowaveJournal, pp. 339-344, May 1987.
7. Bayard, J-P. R., M.E. Cooley an D.H. Schaubert, "Analysis of Infinite Arrays of PrintedDipoles on Dielectric Sheets Perpendicular to a Ground Plane," IEEE Trans. Antennas Propagat.,AP-39, pp. 1722-1732, Dec. 1991.
8. Bayard, J-P.R., M.E. Cooley, and D.H. Schaubert, "Effects of E-Plane Electric Walls onInfinite Arrays of Dipoles Printed on Protruding Dielectric Substrates," IEEE Antennas andPropagation 1992 International Symposium Digest, pp. 1410-1413, Jul. 1992.
9. Schuman, H.K., D.R. Pflug, and L.D. Thompson, "Infinite Phased Arrays of Arbitrarily BentThin Wire Radiators," IEEE Trans. Antennas Propagat., AP-32, pp. 364-377, Apr. 1984.
10. Carver, K.R. and J.W. Mink, "Microstrip Antenna Technology," IEEE Trans. AntennasPropagat., AP-29, pp. 2-24, Jan. 1981.
11. Herd, 1.S., "Full Wave Analysis of Proximity Coupled Rectangular Microstrip AntennaArrays," Electromagnetics, 11, pp. 21-46, Mar. 1991.
12. Jin, J-M. and V.V. Liepa, "Application of Hybrid Finite Element Method to ElectromagneticScattering from Coated Cylinders," IEEE Trans. Antennas Propagat., AP-36, pp. 50-54, Jan.1988.
13. Boyse, W.E. and A.A. Seidl, "A Hybrid Finite Element and Moment Method for Elec-tromagnetic Scattering from Inhomogeneous Objects," Conference on Applied ComputationalElectromagnetics, Monterey, CA, pp. 160-169, Mar. 1991.
128
14. Yuan, Y., D.R. Lynch and J.W. Strohbehn, "Coupling of Finite Element and MomentMethods for Electromagnetic Scattering from Inhomogeneous Objects," IEEE Trans. AntennasPropagat., AP-38, pp. 386-393, Mar. 1990.
15. Morgan, M.A., "Principles of Finite Methods in Electromagnetic Scattering," Progress inElectromagnetics Research: Finite Element and Finite Difference Methods in ElectromagneticScattering, pp. 1-68, New York, NY: Elsevier, 1990.
16. Gedney, S.D., J.F. Lee and R. Mittra, "A Combined FEM/MoM Approach to Analyze thePlane Wave Diffraction by Arbitrary Gratings," IEEE Trans. Microwave Theory Tech., MTT-40,pp. 363-370, Feb. 1992.
17. McGrath, D.T., "Hybrid Finite Element/Waveguide Mode Analysis of Passive RF Devices,"RL-TR-93-, Hanscom AFB ,MA: USAF Rome Laboratory, Feb. 1993.
18. Strang, G. and G.J. Fix, An Analysis of the Finite Element Method, Englewood Cliffs, NJ:Prentice-Hall, 1973.
19. Mur, G., "Finite-Element Modeling of Three-Dimensional Electromagnetic Fields in Inho-mogeneous Media," Radio Science, 26, pp. 275-280, 1991.
20. Paulsen, K.D. and D.R. Lynch, "Elimination of Vector Parasites in Finite Element MaxwellSolutions," IEEE Trans. Microwave Theory Tech., MTT-39, pp. 395-404, Mar. 1991.
21. Boyse, W.E., D.R. Lynch, K.D. Paulsen and G.N. Minerbo, "Nodal-Based Finite ElementModeling of Maxwell's Equations," IEEE Trans. Antennas Propagat., AP-40, pp. 642-651, Jun.1992.
22. Nedelec, J.C., "Mixed Finite Elements in R3," Numerische Mathematik, 35, pp. 315-341,1980.
23. Barton, M.L. and Z.J. Cendes, "New Vector Finite Elements for Three-Dimensional Mag-netic Field Computation," J. Appl. Phys. 61, pp. 3919-3921, Apr. 1987.
24. D'Angelo, J.D. and I.D. Mayergoyz, "Finite Element Methods for the Solution of RFRadiation and Scattering Problems," Electromagnetics, 10, pp. 177-199, 1990.
25. K.D. Paulsen, W.E. Boyse and D.R. Lynch, "Continuous Potential Maxwell Solutions onNodal-Based Finite Elements," IEEE Trans. Antennas Propagat., AP-40, pp. 1192-1200, Oct.1992.
26. Silvester, P.P. and R.L. Ferrari, Finite Elements for Electrical Engineers, 2nd. ed., Cam-bridge Univ. Press, 1990.
27. Harrington, R.F. and J.R. Mautz, "A Generalized Network Formulation for ApertureProblems," IEEE Trans. Antennas Propagat., AP-24, pp. 870-873, Nov. 1976.
129
28. Marcuvitz, N., Waveguide Handbook, New York: McGraw-Hill, 1951.
29. Amitay, N., V. Galindo and C. Wu, Theory and Analysis of Phased Array Antennas, NewYork: Wiley, 1972.
30. Gaskill, J.D., Linear Systems, Fourier Transforms, and Optics, New York: John Wiley &Sons, 1978.
31. IMSL, Inc., Users' Manual: IMSL MA TH/Library - FORTRAN Subroutines for Mathe-matical Applications, Dec. 1989.
32. Anderson, E., et. al., LAPACK Users' Guide, Philadelphia, PA: SIAM, 1992.
33. Sarkar, T.K. and E. Arvas, "On a Class of Finite Step Iterative Methods (Conjugate Direc-tions) for the Solution of an Operator Equation Arising in Electromagnetics," IEEE Trans.Antennas Propagat., AP-33, pp. 1058-1066, Oct. 1985.
34. Potter, P.D. and A.C. Ludwig, "Beamshaping by Use of Higher Order Modes in ConicalHorns," Electromagnetic Horn Antennas, ed. A.W. Love, pp. 203-204, New York: IEEE Press,1976.
35. Masterman, P.H. and P.J.B. Clarricoats, "Computer Field-Matching Solution of WaveguideTransverse Discontinuities," Proc. lEE, 188, pp. 51-63, Jan. 1971.
36. Kowalski, G. and R. Pregla, "Dispersion Characteristics of Shielded Microstrips with FiniteThickness," Arch. Elek. Ubertragung, 25, pp. 193-196, Apr. 1971.
37. Wheeler, H.A., "Transmission Line Properties of a Strip on a Dielectric Sheet on a Plane,"IEEE Trans. Microwave Theory Tech., MTF-25, pp. 631-647, Aug. 1977.
38. Rowe, D.A. and B.Y. Lao, "Numerical Analysis of Shielded Coplanar Waveguide," IEEETrans. Microwave Theory Tech., MTT-31, pp. 911-915, Nov. 1983.
39. Webb, J.P., G.L. Maile, and R.L. Ferrari, "Finite Element Solution of Three DimensionalElectromagnetic Problems," Proc. lEE, 130, pp. 153-159, Mar. 1983.
40. Diamond, B.L., "Resonance Phenomena in Waveguide Arrays," Proc. 1967 IEEE AntennasPropagat. Int'l Symp., Ann Arbor, MI, pp. 110-115, Oct. 1967.
41. Lee, S-W and W. Jones, "On the Suppression of Radiation Nulls and Broadband ImpedanceMatching of Rectangular Waveguide Phased Arrays," IEEE Trans. Antennas Propagat., AP-19,pp. 41-51, Jan. 1971.
42. Amitay, N. and M.J. Gans, "Design of Rectangular Horn Arrays with Oversized ApertureElements," IEEE Trans. Antennas Propagat., AP-29, pp. 871-884, Nov. 1981.
130
43. Tang, R. and N.S. Wong, "Multimode Phased Array Element for Wide Scan Angle Im-pedance Matching," Proc. IEEE, 56, pp. 1951-1959, Nov. 1968.
44. Structural Dynamics Research Corporation, "Integrated Design Engineering AnalysisSoftware Users' Manual," S.D.R.C., Milford, OH, 1990.
45. Pozar, D.M. and D.H. Schaubert, "Analysis of an Infinite Array of Rectangular MicrostripPatches with Idealized Probe Feeds," IEEE Trans. Antennas Propagat., AP-32, pp. 1101-1107,Oct. 1984.
46. Fenn, A.J., "Theoretical and Experimental Study of Monopole Phased Array Antennas,"IEEE Trans. Antennas Propagat., AP-33, pp. 1131-1142, Oct. 1985.
47. Herper, J.C. and A. Hessel, "Performance of X/4 Monopole in a Phased Array," Proc. 1975Antennas and Propagation Int'l Symp., Urbana, IL, Jun. 1975.
48. Hammerstad, E.O., "Equations for Microstrip Circuit Design," Proc. 5th European Micro-waves Conf., Hamburg, pp. 268-272, 1975.
49. Gupta, K.C., R. Gharg, and R. Chadha, Computer-Aided Design of Microwave Circuits,Dedham, MA: Artech House, 1981.
50. Schaubert, D.H., "Endfire Slotline Antennas," Proc. JINA '90, 253-265, Nov. 1990.
St. Choung, Y.H. and C.C. Chen, "44 GHz Slotline Phased Array Antenna," Proc. 1989Antennas and Propagation Int'l Symp., pp. 1730-1733.
52. Ho, T.Q. and S.M. Hart, "A Broad-Band Coplanar Waveguide to Slotline Transition," IEEEMicrowave and Guided Wave Letters, 2, pp. 415-416, Oct. 1992.
53. Jin, J-M. and J.L. Volakis, "A Hybrid Finite Element Method for Scattering and Radiationby Microstrip Patch Antennas and Arrays Residing in a Cavity," IEEE Trans. Antennas Prop-agat, AP-39, pp. 1598-1604, Nov. 1991.
54. Mikhlin, S.G., Variational Methods in Mathematical Physics, New York: MacMillan, 1974.
55. Harrington, R.F., lime-Harmonic Electromagnetic Fields, New York: McGraw-Hill, 1961.
56. Naylor, A.W. and G.R. Sell, Linear Operator Theory in Engineering and Science, NewYork: Springer-Verlag, 1982.
57. Stroud, A.H. and D. Secrest, Gaussian Quadrature Formulas, New York: Prentice-Hall,1966.
58. Abramowitz, M. and I.E. Stegun, Handbook of Mathematical Functions, National Bureauof Standards, 1972.
131
59. Collin, R.E., Field Theory of Guided Waves, 2nd Ed., IEEE Press, 1991.
60. Goodman, J.W., Introduction to Fourier Optics, McGraw-Hill, 1968.
61. Brillouin, L., Wave Propagation in Periodic Structures, New York: Dover, 1953.
62. Mittra, R., C.H. Chan and T. Cwik, "Techniques for Analyzing Frequency Selective Surfac-es--A Review, IEEE Proceedings, 76, pp. 1593-1615, Dec, 1988.
132
Vita
Major Daniel Timothy McGrath was born in Tampa, Florida on October 24, 1956. He
graduated from William J. Palmer High School in Colorado Springs, Colorado in 1974. After
attending the University of Colorado at Colorado Springs for one year, he entered the United
States Air Force Academy, graduating in May, 1979 with the Bachelor of Science in Electrical
Engineering degree. He then worked for two years in the Air Force Armament Laboratory at
Eglin AFB, Florida in the development of sensors, signal processing and pattern recognition for
smart munitions. He entered the Air Force Institute of Technology (AFIT) in June 1981 and
graduated in December, 1982 with the Master of Science in Electrical Engineering. His next
assignment was in the Antennas Division of Rome Air Development Center, Hanscom AFB,
Massachusetts, where he participated in the development of new designs and concepts for phased
array and lens antennas. He entered the Doctor of Science program at AFIT in July, 1990.
After leaving AFIT he will join the Advanced Weapons and Survivability Directorate of Phillips
Laboratory, Kirtland AFB, New Mexico.
133
June 1993 Doctoral Dissertation
PHASED ARRAY ANTENNA ANALYSIS USINGHYBRID FINITE ELEMENT METHODS
Daniel T. McGrath, Major, USAF
Air Force Institute of Technology AFIT/DS/ENG/93-4Wright-Patterson AFB, OH 45433-6583
Rome LaboratoryRL/ERAHanscom AFB, MA 01731
Approved for Public Release;Distribution Unlimited
This research in computational electromagnetics developed a newmethod for predicting the near-field mutual coupling effects inphased array antennas, using the finite element method (FEM) in com-bination with integral equations. Accurate feed modeling is accom-plished by enforcing continuity between the FEM solution and an arbi-trary number of waveguide modes across a ground plane aperture. Aperiodic integral equation is imposed above the antenna's physicalstructure in order to enforce the radiation condition and to confinethe analysis to an array unit cell. The electric field is expandedin terms of vector finite elements, and Galerkin's method is used towrite the problem as a matrix equation. A general-purpose computercode was developed and validated by comparing its results to pub-lished data for several array types. Its versatility was demonstrat-ed with predictions of the scanning properties of arrays of printeddipoles and printed flared notches.
Phased Arrays Finite Element AnalysisMethod of Moments Electromagnetic RadiationBroadband Antennas Microwave Components
Unclassified Unclassified Unclassified UL